pudl.transform.ferc1

Classes & functions to process FERC Form 1 data before loading into the PUDL DB.

Note that many of the classes/objects here inherit from/are instances of classes defined in pudl.transform.classes. Their design and relationships to each other are documented in that module.

See pudl.transform.params.ferc1 for the values that parameterize many of these transformations.

Module Contents

Classes

SourceFerc1	Enumeration of allowed FERC 1 raw data sources.
TableIdFerc1	Enumeration of the allowed FERC 1 table IDs.
RenameColumnsFerc1	Dictionaries for renaming either XBRL or DBF derived FERC 1 columns.
WideToTidy	Parameters for converting a wide table to a tidy table with value types.
WideToTidySourceFerc1	Parameters for converting either or both XBRL and DBF table from wide to tidy.
MergeXbrlMetadata	Parameters for merging in XBRL metadata.
DropDuplicateRowsDbf	Parameter for dropping duplicate DBF rows.
AlignRowNumbersDbf	Parameters for aligning DBF row numbers with metadata from mannual maps.
SelectDbfRowsByCategory	Parameters for select_dbf_rows_by_category().
UnstackBalancesToReportYearInstantXbrl	Parameters for unstack_balances_to_report_year_instant_xbrl().
CombineAxisColumnsXbrl	Parameters for combine_axis_columns_xbrl().
ReconcileTableCalculations	Parameters for reconciling xbrl-metadata based calculations within a table.
Ferc1TableTransformParams	A model defining what TransformParams are allowed for FERC Form 1.
Ferc1AbstractTableTransformer	An abstract class defining methods common to many FERC Form 1 tables.
FuelFerc1TableTransformer	A table transformer specific to the fuel_ferc1 table.
PlantsSteamFerc1TableTransformer	Transformer class for the plants_steam_ferc1 table.
PlantsHydroFerc1TableTransformer	A table transformer specific to the plants_hydro_ferc1 table.
PlantsPumpedStorageFerc1TableTransformer	Transformer class for plants_pumped_storage_ferc1 table.
PurchasedPowerFerc1TableTransformer	Transformer class for purchased_power_ferc1 table.
PlantInServiceFerc1TableTransformer	A transformer for the plant_in_service_ferc1 table.
PlantsSmallFerc1TableTransformer	A table transformer specific to the plants_small_ferc1 table.
TransmissionStatisticsFerc1TableTransformer	A table transformer for the transmission_statistics_ferc1 table.
ElectricEnergySourcesFerc1TableTransformer	Transformer class for electric_energy_sources_ferc1 table.
ElectricEnergyDispositionsFerc1TableTransformer	Transformer class for electric_energy_dispositions_ferc1 table.
UtilityPlantSummaryFerc1TableTransformer	Transformer class for utility_plant_summary_ferc1 table.
BalanceSheetLiabilitiesFerc1TableTransformer	Transformer class for balance_sheet_liabilities_ferc1 table.
BalanceSheetAssetsFerc1TableTransformer	Transformer class for balance_sheet_assets_ferc1 table.
IncomeStatementFerc1TableTransformer	Transformer class for the income_statement_ferc1 table.
RetainedEarningsFerc1TableTransformer	Transformer class for retained_earnings_ferc1 table.
DepreciationAmortizationSummaryFerc1TableTransformer	Transformer class for depreciation_amortization_summary_ferc1 table.
ElectricPlantDepreciationChangesFerc1TableTransformer	Transformer class for electric_plant_depreciation_changes_ferc1 table.
ElectricPlantDepreciationFunctionalFerc1TableTransformer	Transformer for electric_plant_depreciation_functional_ferc1 table.
ElectricOperatingExpensesFerc1TableTransformer	Transformer class for electric_operating_expenses_ferc1 table.
ElectricOperatingRevenuesFerc1TableTransformer	Transformer class for electric_operating_revenues_ferc1 table.
CashFlowFerc1TableTransformer	Transform class for cash_flow_ferc1 table.
ElectricitySalesByRateScheduleFerc1TableTransformer	Transform class for electricity_sales_by_rate_schedule_ferc1 table.
OtherRegulatoryLiabilitiesFerc1TableTransformer	Transformer class for other_regulatory_liabilities_ferc1 table.

Functions

clean_xbrl_metadata_json(→ dict[str, dict[str, ...)	Generate cleaned json xbrl metadata.
add_source_tables_to_xbrl_metadata(→ dict[str, ...)	Add a source_tables field into metadata calculation components.
wide_to_tidy(→ pandas.DataFrame)	Reshape wide tables with FERC account columns to tidy format.
merge_xbrl_metadata(→ pandas.DataFrame)	Merge metadata based on params.
drop_duplicate_rows_dbf(→ pandas.DataFrame)	Drop duplicate DBF rows if duplicates have indentical data or one row has nulls.
align_row_numbers_dbf(→ pandas.DataFrame)	Rename the xbrl_factoid column after align_row_numbers_dbf().
select_dbf_rows_by_category(→ pandas.DataFrame)	Select DBF rows with values listed or found in XBRL in a categorical-like column.
unstack_balances_to_report_year_instant_xbrl(...)	Turn start year end year rows into columns for each value type.
combine_axis_columns_xbrl(→ pandas.DataFrame)	Combine axis columns from squished XBRL tables into one column with no NAs.
reconcile_table_calculations(→ pandas.DataFrame)	Ensure intra-table calculated values match reported values within a tolerance.
calculate_values_from_components(→ pandas.DataFrame)	Apply calculations derived from XBRL metadata to reported XBRL data.
check_calculation_metrics(→ pandas.DataFrame)	Run the calculation metrics and determine if calculations are within tolerance.
get_ferc1_dbf_rows_to_map(→ pandas.DataFrame)	Identify DBF rows that need to be mapped to XBRL columns.
update_dbf_to_xbrl_map(→ pandas.DataFrame)	Regenerate the FERC 1 DBF+XBRL glue while retaining existing mappings.
read_dbf_to_xbrl_map(→ pandas.DataFrame)	Read the manually compiled DBF row to XBRL column mapping for a given table.
fill_dbf_to_xbrl_map(→ pandas.DataFrame)	Forward-fill missing years in the minimal, manually compiled DBF to XBRL mapping.
get_data_cols_raw_xbrl(→ list[str])	Get a list of all XBRL data columns appearing in a given XBRL table.
read_xbrl_calculation_fixes(→ pandas.DataFrame)	Read in the table of calculation fixes.
ferc1_transform_asset_factory(→ dagster.AssetsDefinition)	Create an asset that pulls in raw ferc Form 1 assets and applies transformations.
create_ferc1_transform_assets(...)	Create a list of transformed FERC Form 1 assets.
plants_steam_ferc1(→ pandas.DataFrame)	Create the clean plants_steam_ferc1 table.
other_dimensions(→ list[str])	Get a list of the other dimension columns across all of the transformers.
table_to_xbrl_factoid_name(→ dict[str, str])	Build a dictionary of table name (keys) to xbrl_factiod column name.
table_dimensions_ferc1(→ pandas.DataFrame)	Build a table of values of dimensions observed in the transformed data tables.
metadata_xbrl_ferc1(→ pandas.DataFrame)	Build a table of all of the tables' XBRL metadata.
calculation_components_xbrl_ferc1(→ pandas.DataFrame)	Create calculation-compnent table from table-level metadata.
unexpected_total_components(→ pandas.DataFrame)	Find unexpected components in within-fact total calculations.
check_for_calc_components_duplicates(→ None)	Check for duplicates calculation records.
make_calculation_dimensions_explicit(→ pandas.DataFrame)	Fill in null dimensions w/ the values observed in table_dimensions_ferc1().
assign_parent_dimensions(→ pandas.DataFrame)	Add dimensions to calculation parents.
infer_intra_factoid_totals(→ pandas.DataFrame)	Define dimension total calculations.

Attributes

logger
FERC1_TFR_CLASSES
ferc1_assets

pudl.transform.ferc1.logger

pudl.transform.ferc1.clean_xbrl_metadata_json(raw_xbrl_metadata_json: dict[str, dict[str, list[dict[str, Any]]]]) → dict[str, dict[str, list[dict[str, Any]]]]

Generate cleaned json xbrl metadata.

For now, this only runs add_source_tables_to_xbrl_metadata().

pudl.transform.ferc1.add_source_tables_to_xbrl_metadata(raw_xbrl_metadata_json: dict[str, dict[str, list[dict[str, Any]]]]) → dict[str, dict[str, list[dict[str, Any]]]]

Add a source_tables field into metadata calculation components.

When a particular component of a calculation does not originate from the table in which the calculated field is being reported, label the source table.

class pudl.transform.ferc1.SourceFerc1(*args, **kwds)

Bases: enum.Enum

Enumeration of allowed FERC 1 raw data sources.

XBRL = 'xbrl'

DBF = 'dbf'

class pudl.transform.ferc1.TableIdFerc1(*args, **kwds)

Bases: enum.Enum

Enumeration of the allowed FERC 1 table IDs.

Hard coding this doesn’t seem ideal. Somehow it should be either defined in the context of the Package, the Ferc1Settings, an etl_group, or DataSource. All of the table transformers associated with a given data source should have a table_id that’s from that data source’s subset of the database. Where should this really happen? Alternatively, the allowable values could be derived from the structure of the Package. But this works for now.

FUEL_FERC1 = 'fuel_ferc1'

PLANTS_STEAM_FERC1 = 'plants_steam_ferc1'

PLANTS_HYDRO_FERC1 = 'plants_hydro_ferc1'

PLANTS_SMALL_FERC1 = 'plants_small_ferc1'

PLANTS_PUMPED_STORAGE_FERC1 = 'plants_pumped_storage_ferc1'

PLANT_IN_SERVICE_FERC1 = 'plant_in_service_ferc1'

PURCHASED_POWER_FERC1 = 'purchased_power_ferc1'

TRANSMISSION_STATISTICS_FERC1 = 'transmission_statistics_ferc1'

ELECTRIC_ENERGY_SOURCES_FERC1 = 'electric_energy_sources_ferc1'

ELECTRIC_ENERGY_DISPOSITIONS_FERC1 = 'electric_energy_dispositions_ferc1'

UTILITY_PLANT_SUMMARY_FERC1 = 'utility_plant_summary_ferc1'

ELECTRIC_OPERATING_EXPENSES_FERC1 = 'electric_operating_expenses_ferc1'

BALANCE_SHEET_LIABILITIES = 'balance_sheet_liabilities_ferc1'

DEPRECIATION_AMORTIZATION_SUMMARY_FERC1 = 'depreciation_amortization_summary_ferc1'

BALANCE_SHEET_ASSETS_FERC1 = 'balance_sheet_assets_ferc1'

RETAINED_EARNINGS_FERC1 = 'retained_earnings_ferc1'

INCOME_STATEMENT_FERC1 = 'income_statement_ferc1'

ELECTRIC_PLANT_DEPRECIATION_CHANGES_FERC1 = 'electric_plant_depreciation_changes_ferc1'

ELECTRIC_OPERATING_REVENUES_FERC1 = 'electric_operating_revenues_ferc1'

ELECTRIC_PLANT_DEPRECIATION_FUNCTIONAL_FERC1 = 'electric_plant_depreciation_functional_ferc1'

CASH_FLOW_FERC1 = 'cash_flow_ferc1'

ELECTRICITY_SALES_BY_RATE_SCHEDULE_FERC1 = 'electricity_sales_by_rate_schedule_ferc1'

OTHER_REGULATORY_LIABILITIES_FERC1 = 'other_regulatory_liabilities_ferc1'

class pudl.transform.ferc1.RenameColumnsFerc1

Bases: pudl.transform.classes.TransformParams

Dictionaries for renaming either XBRL or DBF derived FERC 1 columns.

This is FERC 1 specific, because we need to store both DBF and XBRL rename dictionaires separately. Note that this parameter model does not have its own unique transform function. Like the generic pudl.transform.classes.RenameColumns it depends on the build in pd.rename() method, which is called with the values DBF or XBRL parameters depending on the context.

Potential parameters validations that could be implemented

• Validate that all keys appear in the original dbf/xbrl sources. This has to be true, but right now we don’t have stored metadata enumerating all of the columns that exist in the raw data, so we don’t have anything to check against. Implement once when we have schemas defined for after the extract step.

• Validate all values appear in PUDL tables, and all expected PUDL names are mapped. Actually we can’t require that the rename values appear in the PUDL tables, because there will be cases in which the original column gets dropped or modified, e.g. in the case of unit conversions with a column rename.

property rename_dicts_xbrl

Compile all of the XBRL rename dictionaries into an ordered list.

dbf: pudl.transform.classes.RenameColumns

xbrl: pudl.transform.classes.RenameColumns

duration_xbrl: pudl.transform.classes.RenameColumns

instant_xbrl: pudl.transform.classes.RenameColumns

class pudl.transform.ferc1.WideToTidy

Bases: pudl.transform.classes.TransformParams

Parameters for converting a wide table to a tidy table with value types.

idx_cols: list[str] | None

List of column names to treat as the table index.

stacked_column_name: str | None

Name of column that will contain the stacked categories.

value_types: list[str] | None

List of names of value types that will end up being the column names.

Some of the FERC tables have multiple data types spread across many different categories. In the input dataframe given to wide_to_tidy(), the value types must be the suffixes of the column names. If the table does not natively have the pattern of “{to-be stacked category}_{value_type}”, rename the columns using a rename_columns.duration_xbrl, rename_columns.instant_xbrl or rename_columns.dbf parameter which will be employed in process_duration_xbrl(), process_instant_xbrl() or process_dbf().

expected_drop_cols: int = 0

The number of columns that are expected to be dropped.

wide_to_tidy_xbrl() will generate a regex pattern assuming the value_types are the column name’s suffixes. If a column does not conform to that pattern, it will be filtered out. This is helpful for us to not include a bunch of columns from the input dataframe incorrectly included in the stacking process. We could enumerate every column that we want to drop, but this could be tedious and potentially error prone. But this does mean that if a column is incorrectly named - or barely missing the pattern, it will be dropped. This parameter enables us to lock the number of expected columns. If the dropped columns are a different number, an error will be raised.

class pudl.transform.ferc1.WideToTidySourceFerc1

Bases: pudl.transform.classes.TransformParams

Parameters for converting either or both XBRL and DBF table from wide to tidy.

property value_types: list[str]

Compile a list of all of the value_types from wide_to_tidy.

xbrl: WideToTidy | list[WideToTidy]

dbf: WideToTidy | list[WideToTidy]

pudl.transform.ferc1.wide_to_tidy(df: pandas.DataFrame, params: WideToTidy) → pandas.DataFrame

Reshape wide tables with FERC account columns to tidy format.

The XBRL table coming into this method could contain all the data from both the instant and duration tables in a wide format – with one column for every combination of value type (e.g. additions, ending_balance) and value category, which means ~500 columns for some tables.

We tidy this into a long table with one column for each of the value types in params.value_types and a new column named xbrl_factoid that contains categories that were previously the XBRL column name stems.

This allows aggregations of multiple xbrl_factoid categories in a columnar fashion such as aggregation across groups of rows to total up various hierarchical accounting categories (hydraulic turbines -> hydraulic production plant -> all production plant -> all electric utility plant) though the categorical columns required for that aggregation are added later.

For table that have a internal relationship between the values in the params.value_types, such as the plant_in_service_ferc1 table, this also enables aggregation across columns to calculate the ending balance based on the starting balance and all of the reported changes.

class pudl.transform.ferc1.MergeXbrlMetadata

Bases: pudl.transform.classes.TransformParams

Parameters for merging in XBRL metadata.

rename_columns: dict[str, str]

Dictionary to rename columns in the normalized metadata before merging.

This dictionary will be passed as pd.DataFrame.rename() columns parameter.

on: str | None

Column name to merge on in merge_xbrl_metadata().

pudl.transform.ferc1.merge_xbrl_metadata(df: pandas.DataFrame, xbrl_metadata: pandas.DataFrame, params: MergeXbrlMetadata) → pandas.DataFrame

Merge metadata based on params.

class pudl.transform.ferc1.DropDuplicateRowsDbf

Bases: pudl.transform.classes.TransformParams

Parameter for dropping duplicate DBF rows.

table_name: TableIdFerc1 | None

Name of table used to grab primary keys of PUDL table to check for duplicates.

data_columns: list = []

List of data column names to ensure primary key duplicates have the same data.

pudl.transform.ferc1.drop_duplicate_rows_dbf(df: pandas.DataFrame, params: DropDuplicateRowsDbf, return_dupes_w_unique_data: bool = False) → pandas.DataFrame

Drop duplicate DBF rows if duplicates have indentical data or one row has nulls.

There are several instances of the DBF data reporting the same value on multiple rows. This function checks to see if all of the duplicate values that have the same primary keys have reported the same data or have records with null data in any of the data columns while the other record has complete data. If the duplicates have no unique data, the duplicates are dropped with keep="first". If any duplicates do not contain the same data or half null data, an assertion will be raised.

Parameters:

• df – DBF table containing PUDL primary key columns

• params – an instance of DropDuplicateRowsDbf

• return_dupes_w_unique_data – Boolean flag used for debuging only which returns the duplicates which contain actually unique data instead of raising assertion. Default is False.

class pudl.transform.ferc1.AlignRowNumbersDbf

Bases: pudl.transform.classes.TransformParams

Parameters for aligning DBF row numbers with metadata from mannual maps.

dbf_table_names: list[str] | None

DBF table to use to grab the row map in align_row_numbers_dbf().

Default is None.

pudl.transform.ferc1.align_row_numbers_dbf(df: pandas.DataFrame, params: AlignRowNumbersDbf) → pandas.DataFrame

Rename the xbrl_factoid column after align_row_numbers_dbf().

class pudl.transform.ferc1.SelectDbfRowsByCategory

Bases: pudl.transform.classes.TransformParams

Parameters for select_dbf_rows_by_category().

column_name: str | None

The column name containing categories to select by.

select_by_xbrl_categories: bool = False

Boolean flag to indicate whether or not to use the categories in the XBRL table.

If True, select_dbf_rows_by_category() will find the list of categories that exist in the passed in processed_xbrl to select by.

additional_categories: list[str] = []

List of additional categories to select by.

If select_by_xbrl_categories is True, these categories will be added to the XBRL categories and both will be used to select rows from the DBF data. If select_by_xbrl_categories is False, only the “additional” categories will be the used to select rows from the DBF data.

len_expected_categories_to_drop: int = 0

Number of categories that are expected to be dropped from the DBF data.

This is here to ensure no unexpected manipulations to the categories have occured. A warning will be flagged if this number is different than the number of categories that are being dropped.

pudl.transform.ferc1.select_dbf_rows_by_category(processed_dbf: pandas.DataFrame, processed_xbrl: pandas.DataFrame, params: SelectDbfRowsByCategory) → pandas.DataFrame

Select DBF rows with values listed or found in XBRL in a categorical-like column.

The XBRL data often breaks out sub-sections of DBF tables into their own table. These breakout tables are often messy, unstructured portions of a particular schedule or page on the FERC1 PDF. We often want to preserve some of the ways the XBRL data is segmented so we need to be able to select only portions of the DBF table to be concatenated with the XBRL data.

In mapping DBF data to XBRL data for the tables that rely on their row_number we map each row to its corresponding xbrl_factoid. The standard use of this transformer is to use the column_name that corresponds to the xbrl_factoid that was merged into the DBF data via align_row_numbers_dbf() and was converted into a column in the XBRL data via wide_to_tidy().

Note: Often, the unstructured portion of the DBF table that (possibly) sums up into a single value in structured data has the same xbrl_factoid name in the XBRL tables. By convention, we are employing a pattern in the dbf_to_xbrl.csv map that involves adding an _unstructed suffix to the rows that correspond to the unstructured portion of the table. This enables a simple selection of the structured part of the table. When processing the unstructured table, you can either rename the XBRL data’s factoid name to include an _unstructed suffix or you can specify the categories with _unstructed suffixes using the additional_categories parameter.

class pudl.transform.ferc1.UnstackBalancesToReportYearInstantXbrl

Bases: pudl.transform.classes.TransformParams

Parameters for unstack_balances_to_report_year_instant_xbrl().

unstack_balances_to_report_year: bool = False

If True unstack balances to a single year (the report year).

pudl.transform.ferc1.unstack_balances_to_report_year_instant_xbrl(df: pandas.DataFrame, params: UnstackBalancesToReportYearInstantXbrl, primary_key_cols: list[str]) → pandas.DataFrame

Turn start year end year rows into columns for each value type.

Called in Ferc1AbstractTableTransformer.process_instant_xbrl().

Some instant tables report year-end data, with their datestamps in different years, but we want year-start and year-end data within a single report_year (which is equivalent) stored in two separate columns for compatibility with the DBF data.

This function unstacks that table and adds the suffixes _starting_balance and _ending_balance to each of the columns. These can then be used as value_types in wide_to_tidy() to normalize the table.

There are two checks in place:

First, it will make sure that there are not duplicate entries for a single year + other primary key fields. Ex: a row for 2020-12-31 and 2020-06-30 for entitiy_id X means that the data isn’t annually unique. We could just drop these mid-year values, but we might want to keep them or at least check that there is no funny business with the data.

We also check that there are no mid-year dates at all. If an entity reports a value from the middle of the year, we can’t identify it as a start/end of year value.

Params:

primary_key_cols: The columns that should be used to check for duplicated data,

and also for unstacking the balance – these are set to be the index before unstack is called. These are typically set by the wrapping method and generated automatically based on other class transformation parameters via Ferc1AbstractTableTransformer.source_table_primary_key().

class pudl.transform.ferc1.CombineAxisColumnsXbrl

Bases: pudl.transform.classes.TransformParams

Parameters for combine_axis_columns_xbrl().

axis_columns_to_combine: list | None

List of axis columns to combine.

new_axis_column_name: str | None

The name of the combined axis column – must end with the suffix _axis!.

doesnt_end_with_axis(v)

Ensure that new axis column ends in _axis.

pudl.transform.ferc1.combine_axis_columns_xbrl(df: pandas.DataFrame, params: CombineAxisColumnsXbrl) → pandas.DataFrame

Combine axis columns from squished XBRL tables into one column with no NAs.

Called in Ferc1AbstractTableTransformer.process_xbrl().

There are instances (ex: sales_by_rate_schedule_ferc1) where the DBF table is equal to several concatenated XBRL tables. These XBRL tables are extracted together with the function extract_xbrl_concat(). Once combined, we need to deal with their axis columns.

We use the axis columns (the primary key for the raw XBRL tables) in the creation of record_id``s for each of the rows. If each of the concatinated XBRL tables has the same axis column name then there's no need to fret. However, if the columns have slightly different names (ex: ``residential_sales_axis vs. industrial_sales_axis), we’ll need to combine them. We combine them to get rid of NA values which aren’t allowed in primary keys. Otherwise it would look like this:

residential_sales_axis	industrial_sales_axis
value1	NA
value2	NA
NA	valueA
NA	valueB

vs. this:

sales_axis
value1
value2
valueA
valueB

class pudl.transform.ferc1.ReconcileTableCalculations

Bases: pudl.transform.classes.TransformParams

Parameters for reconciling xbrl-metadata based calculations within a table.

column_to_check: str | None

Name of data column to check.

This will typically be dollar_value or ending_balance column for the income statement and the balance sheet tables.

calculation_tolerance: float = 0.05

Fraction of calculated values which we allow not to match reported values.

subtotal_column: str | None

Sub-total column name (e.g. utility type) to compare calculations against in reconcile_table_calculations().

subtotal_calculation_tolerance: float = 0.05

Fraction of calculated sub-totals allowed not to match reported values.

pudl.transform.ferc1.reconcile_table_calculations(df: pandas.DataFrame, calculation_components: pandas.DataFrame, xbrl_factoid_name: str, table_name: str, params: ReconcileTableCalculations, add_corrections: bool = True) → pandas.DataFrame

Ensure intra-table calculated values match reported values within a tolerance.

In addition to checking whether all reported “calculated” values match the output of our repaired calculations, this function adds a correction record to the dataframe that is included in the calculations so that after the fact the calculations match exactly. This is only done when the fraction of records that don’t match within the tolerances of numpy.isclose() is below a set threshold.

Note that only calculations which are off by a significant amount result in the creation of a correction record. Many calculations are off from the reported values by exaclty one dollar, presumably due to rounding errrors. These records typically do not fail the numpy.isclose() test and so are not corrected.

Parameters:

• df – processed table.

• calculation_components – processed calculation component metadata.

• xbrl_factoid_name – column name of the XBRL factoid in the processed table.

• table_name – name of the PUDL table.

• params – ReconcileTableCalculations parameters.

• add_corrections – Whether or not to create _correction records that force all calculations to add up correctly.

pudl.transform.ferc1.calculate_values_from_components(calculation_components: pandas.DataFrame, data: pandas.DataFrame, calc_idx: list[str], value_col: str) → pandas.DataFrame

Apply calculations derived from XBRL metadata to reported XBRL data.

Parameters:

• calculation_components – Table defining the calculations, with each row defining a single component, including its weight. Groups of rows identified by table_name_parent and xbrl_factoid_parent indicate the values being calculated.

• data – exploded FERC data to apply the calculations to. Primary key should be report_year, utility_id_ferc1, table_name, xbrl_factoid, and whatever additional dimensions are relevant to the data.

• validate – type of merge validation to apply when initially merging the calculation components (left) and the data (right).

• calc_idx – primary key columns that uniquely identify a calculation component (not including the _parent columns).

• value_col – label of the column in data that contains the values to apply the calculations to (typically dollar_value or ending_balance).

pudl.transform.ferc1.check_calculation_metrics(calculated_df: pandas.DataFrame, value_col: str, calculation_tolerance: float, table_name: str, add_corrections: bool = True) → pandas.DataFrame

Run the calculation metrics and determine if calculations are within tolerance.

class pudl.transform.ferc1.Ferc1TableTransformParams

Bases: pudl.transform.classes.TableTransformParams

A model defining what TransformParams are allowed for FERC Form 1.

This adds additional parameter models beyond the ones inherited from the pudl.transform.classes.AbstractTableTransformer class.

property xbrl_factoid_name: str

Access the column name of the xbrl_factoid.

property rename_dicts_xbrl

Compile all of the XBRL rename dictionaries into an ordered list.

property wide_to_tidy_value_types: list[str]

Compile a list of all of the value_types from wide_to_tidy.

property aligned_dbf_table_names: list[str]

The list of DBF tables aligned by row number in this transform.

rename_columns_ferc1: RenameColumnsFerc1

wide_to_tidy: WideToTidySourceFerc1

merge_xbrl_metadata: MergeXbrlMetadata

align_row_numbers_dbf: AlignRowNumbersDbf

drop_duplicate_rows_dbf: DropDuplicateRowsDbf

select_dbf_rows_by_category: SelectDbfRowsByCategory

unstack_balances_to_report_year_instant_xbrl: UnstackBalancesToReportYearInstantXbrl

combine_axis_columns_xbrl: CombineAxisColumnsXbrl

reconcile_table_calculations: ReconcileTableCalculations

pudl.transform.ferc1.get_ferc1_dbf_rows_to_map(ferc1_engine: sqlalchemy.engine.Engine) → pandas.DataFrame

Identify DBF rows that need to be mapped to XBRL columns.

Select all records in the f1_row_lit_tbl where the row literal associated with a given combination of table and row number is different from the preceeding year. This is the smallest set of records which we can use to reproduce the whole table by expanding the time series to include all years, and forward filling the row literals.

pudl.transform.ferc1.update_dbf_to_xbrl_map(ferc1_engine: sqlalchemy.engine.Engine) → pandas.DataFrame

Regenerate the FERC 1 DBF+XBRL glue while retaining existing mappings.

Reads all rows that need to be mapped out of the f1_row_lit_tbl and appends columns containing any previously mapped values, returning the resulting dataframe.

pudl.transform.ferc1.read_dbf_to_xbrl_map(dbf_table_names: list[str]) → pandas.DataFrame

Read the manually compiled DBF row to XBRL column mapping for a given table.

Parameters:

dbf_table_name – The original name of the table in the FERC Form 1 DBF database whose mapping to the XBRL data you want to extract. for example f1_plant_in_srvce.

Returns:

DataFrame with columns [sched_table_name, report_year, row_number, row_type, xbrl_factoid]

pudl.transform.ferc1.fill_dbf_to_xbrl_map(df: pandas.DataFrame, dbf_years: list[int] | None = None) → pandas.DataFrame

Forward-fill missing years in the minimal, manually compiled DBF to XBRL mapping.

The relationship between a DBF row and XBRL column/fact/entity/whatever is mostly consistent from year to year. To minimize the amount of manual mapping work we have to do, we only map the years in which the relationship changes. In the end we do need a complete correspondence for all years though, and this function uses the minimal information we’ve compiled to fill in all the gaps, producing a complete mapping across all requested years.

One complication is that we need to explicitly indicate which DBF rows have headers in them (which don’t exist in XBRL), to differentiate them from null values in the exhaustive index we create below. We set a HEADER_ROW sentinel value so we can distinguish between two different reasons that we might find NULL values in the xbrl_factoid field:

1. It’s NULL because it’s between two valid mapped values (the NULL was created in our filling of the time series) and should thus be filled in, or

2. It’s NULL because it was a header row in the DBF data, which means it should NOT be filled in. Without the HEADER_ROW value, when a row number from year X becomes associated with a non-header row in year X+1 the ffill will keep right on filling, associating all of the new header rows with the value of xbrl_factoid that was associated with the old row number.

Parameters:

• df – A dataframe containing a DBF row to XBRL mapping for a single FERC 1 DBF table.

• dbf_years – The list of years that should have their DBF row to XBRL mapping filled in. This defaults to all available years of DBF data for FERC 1. In general this parameter should only be set to a non-default value for testing purposes.

Returns:

A complete mapping of DBF row number to XBRL columns for all years of data within a single FERC 1 DBF table. Has columns of [report_year, row_number, xbrl_factoid]

pudl.transform.ferc1.get_data_cols_raw_xbrl(raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → list[str]

Get a list of all XBRL data columns appearing in a given XBRL table.

Returns:

A list of all the data columns found in the original XBRL DB that correspond to the given PUDL table. Includes columns from both the instant and duration tables but excludes structural columns that appear in all XBRL tables.

pudl.transform.ferc1.read_xbrl_calculation_fixes() → pandas.DataFrame

Read in the table of calculation fixes.

class pudl.transform.ferc1.Ferc1AbstractTableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: pudl.transform.classes.AbstractTableTransformer

An abstract class defining methods common to many FERC Form 1 tables.

This subclass remains abstract because it does not define transform_main(), which is always going to be table-specific.

• Methods that only apply to XBRL data should end with _xbrl

• Methods that only apply to DBF data should end with _dbf

table_id: TableIdFerc1

parameter_model

params: Ferc1AbstractTableTransformer.parameter_model

has_unique_record_ids: bool = True

True if each record in the transformed table corresponds to one input record.

For tables that have been transformed from wide-to-tidy format, or undergone other kinds of reshaping, there is not a simple one-to-one relationship between input and output records, and so we should not expect record IDs to be unique. In those cases they serve only a forensic purpose, telling us where to find the original source of the transformed data.

xbrl_metadata: pandas.DataFrame

Dataframe combining XBRL metadata for both instant and duration table columns.

xbrl_calculations: pandas.DataFrame | None

Dataframe of calculation components.

If None, the calculations have not been instantiated. If the table has been instantiated but is an empty table, then there are no calculations for that table.

transform_start(raw_dbf: pandas.DataFrame, raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → pandas.DataFrame

Process the raw data until the XBRL and DBF inputs have been unified.

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Generic FERC1 main table transformer.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

A single transformed table concatenating multiple years of cleaned data derived from the raw DBF and/or XBRL inputs.

transform_end(df: pandas.DataFrame) → pandas.DataFrame

Standardized final cleanup after the transformations are done.

Checks calculations. Enforces dataframe schema. Checks for empty dataframes and null columns.

select_dbf_rows_by_category(processed_dbf: pandas.DataFrame, processed_xbrl: pandas.DataFrame, params: SelectDbfRowsByCategory | None = None) → pandas.DataFrame

Wrapper method for select_dbf_rows_by_category().

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Normalize the XBRL JSON metadata, turning it into a dataframe.

This process concatenates and deduplicates the metadata which is associated with the instant and duration tables, since the metadata is only combined with the data after the instant and duration (and DBF) tables have been merged. This happens in Ferc1AbstractTableTransformer.merge_xbrl_metadata().

process_xbrl_metadata(xbrl_metadata_converted: pandas.DataFrame, xbrl_calculations: pandas.DataFrame) → pandas.DataFrame

Process XBRL metadata after the calculations have been cleaned.

Add row_type_xbrl and is_within_table_calc columns and create xbrl_factoid records for the calculation corrections.

Parameters:

• xbrl_metadata_converted – Dataframe of relatively unprocessed metadata. Result of convert_xbrl_metadata_json_to_df().

• xbrl_calculations – Dataframe of calculation components. Result of process_xbrl_metadata_calculations().

deduplicate_xbrl_factoid_xbrl_metadata(tbl_meta: pandas.DataFrame) → pandas.DataFrame

De-duplicate the xbrl_metadata based on xbrl_factoid.

Default is to do nothing besides check for duplicate values because almost all tables have no deduping. Deduplication needs to be applied before the apply_xbrl_calculation_fixes() inside of process_xbrl_metadata().

raw_xbrl_factoid_to_pudl_name(col_name_xbrl: str) → str

Rename a column name from original XBRL name to the transformed PUDL name.

There are several transform params that either explicitly or implicity rename columns: * RenameColumnsFerc1 * WideToTidySourceFerc1 * UnstackBalancesToReportYearInstantXbrl * ConvertUnits

This method attempts to use the table params to translate a column name.

Note: Instead of doing this for each individual column name, we could compile a rename dict for the whole table with a similar processand then apply it for each group of columns instead of running through this full process every time. If this took longer than… ~5 ms on a single table w/ lots of calcs this would probably be worth it for simplicity.

rename_xbrl_factoid(col: pandas.Series) → pandas.Series

Rename a series of raw to PUDL factoid names via raw_xbrl_factoid_to_pudl_name().

rename_xbrl_factoid_other_tables(calc_comps)

Rename the factoids from calculation components from other tables.

Note: It is probably possible to build an apply style function that takes a series of factoid names and a series of table names and returns a table-specific rename_xbrl_factoid.

static add_metadata_corrections(tbl_meta: pandas.DataFrame) → pandas.DataFrame

Create metadata records for the calculation correction factoids.

Parameters:

tbl_meta – processed metadata table which contains columns row_type_xbrl.

add_calculation_corrections(calc_components: pandas.DataFrame) → pandas.DataFrame

Add correction components and parent-only factoids to calculation metadata.

Parameters:

tbl_meta – Partially transformed table metadata in dataframe form.

Returns:

An updated version of the table metadata containing calculation definitions that include a correction component.

get_xbrl_calculation_fixes() → pandas.DataFrame

Grab the XBRL calculation file.

apply_xbrl_calculation_fixes(calc_components: pandas.DataFrame, calc_fixes: pandas.DataFrame) → pandas.DataFrame

Use the fixes we’ve compiled to update calculations in the XBRL metadata.

Note: Temp fix. These updates should probably be moved into the table params and integrated into the calculations via TableCalcs.

process_xbrl_metadata_calculations(xbrl_metadata_converted: pandas.DataFrame) → pandas.DataFrame

Convert xbrl metadata calculations into a table of calculation components.

This method extracts the calculations from the xbrl_metadata_converted that are stored as json embedded within the calculations``column and convert those into calculation component records. The resulting table includes columns pertaining to both the calculation components and the parent factoid that the components pertain to. The parental columns had suffixes of ``_parent.

This method also adds fixes to the calculations via apply_xbrl_calculation_fixes(), adds corrections records via add_calculation_corrections() and adds the column is_within_table_calc.

Parameters:

xbrl_metadata_converted – Dataframe of relatively unprocessed metadata. Result of convert_xbrl_metadata_json_to_df().

merge_xbrl_metadata(df: pandas.DataFrame, params: MergeXbrlMetadata | None = None) → pandas.DataFrame

Combine XBRL-derived metadata with the data it pertains to.

While the metadata we’re using to annotate the data comes from the more recent XBRL data, it applies generally to all the historical DBF data as well! This method reads the normalized metadata out of an attribute.

align_row_numbers_dbf(df: pandas.DataFrame, params: AlignRowNumbersDbf | None = None) → pandas.DataFrame

Align historical FERC1 DBF row numbers with XBRL account IDs.

Additional Parameterization TBD with additional experience. See: https://github.com/catalyst-cooperative/pudl/issues/2012

drop_duplicate_rows_dbf(df: pandas.DataFrame, params: DropDuplicateRowsDbf | None = None) → pandas.DataFrame

Drop the DBF rows where the PKs and data columns are duplicated.

Wrapper function for drop_duplicate_rows_dbf().

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

DBF-specific transformations that take place before concatenation.

process_xbrl(raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → pandas.DataFrame

XBRL-specific transformations that take place before concatenation.

rename_columns(df: pandas.DataFrame, rename_stage: Literal[dbf, xbrl, xbrl_instant, xbrl_duration] | None = None, params: pudl.transform.classes.RenameColumns | None = None)

Grab the params based on the rename stage and run default rename_columns.

Parameters:

• df – Table to be renamed.

• rename_stage – Name of stage in the transform process. Used to get specific stage’s parameters if None have been passed.

• params – Rename column parameters.

select_annual_rows_dbf(df)

Select only annually reported DBF Rows.

There are some DBF tables that include a mix of reporting frequencies. For now, the default for PUDL tables is to have only the annual records.

unstack_balances_to_report_year_instant_xbrl(df: pandas.DataFrame, params: UnstackBalancesToReportYearInstantXbrl | None = None) → pandas.DataFrame

Turn start year end year rows into columns for each value type.

wide_to_tidy(df: pandas.DataFrame, source_ferc1: SourceFerc1, params: WideToTidy | None = None) → pandas.DataFrame

Reshape wide tables with FERC account columns to tidy format.

The XBRL table coming into this method contains all the data from both the instant and duration tables in a wide format – with one column for every combination of value type (e.g. additions, ending_balance) and accounting category, which means ~500 columns.

We tidy this into a long table with one column for each of the value types (6 in all), and a new column that contains the accounting categories. This allows aggregation across columns to calculate the ending balance based on the starting balance and all of the reported changes, and aggregation across groups of rows to total up various hierarchical accounting categories (hydraulic turbines -> hydraulic production plant -> all production plant -> all electric utility plant) though the categorical columns required for that aggregation are added later.

combine_axis_columns_xbrl(df: pandas.DataFrame, params: CombineAxisColumnsXbrl | None = None) → pandas.DataFrame

Combine axis columns from squished XBRL tables into one column with no NA.

merge_instant_and_duration_tables_xbrl(raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → pandas.DataFrame

Merge XBRL instant and duration tables, reshaping instant as needed.

FERC1 XBRL instant period signifies that it is true as of the reported date, while a duration fact pertains to the specified time period. The date column for an instant fact corresponds to the end_date column of a duration fact.

When merging the instant and duration tables, we need to preserve row order. For the small generators table, row order is how we label and extract information from header and note rows. Outer merging messes up the order, so we need to use a one-sided merge. So far, it seems like the duration df contains all the index values in the instant df. To be sure, there’s a check that makes sure there are no unique intant df index values. If that passes, we merge the instant table into the duration table, and the row order is preserved.

Note: This should always be applied before :meth:rename_columns

Parameters:

• raw_xbrl_instant – table representing XBRL instant facts.

• raw_xbrl_duration – table representing XBRL duration facts.

Returns:

A unified table combining the XBRL duration and instant facts, if both types of facts were present. If either input dataframe is empty, the other dataframe is returned unchanged, except that several unused columns are dropped. If both input dataframes are empty, an empty dataframe is returned.

process_instant_xbrl(df: pandas.DataFrame) → pandas.DataFrame

Pre-processing required to make instant and duration tables compatible.

Column renaming is sometimes required because a few columns in the instant and duration tables do not have corresponding names that follow the naming conventions of ~95% of all the columns, which we rely on programmatically when reshaping and concatenating these tables together.

process_duration_xbrl(df: pandas.DataFrame) → pandas.DataFrame

Pre-processing required to make instant and duration tables compatible.

Column renaming is sometimes required because a few columns in the instant and duration tables do not have corresponding names that follow the naming conventions of ~95% of all the columns, which we rely on programmatically when reshaping and concatenating these tables together.

select_current_year_annual_records_duration_xbrl(df)

Select for annual records within their report_year.

Select only records that have a start_date at begining of the report_year and have an end_date at the end of the report_year.

drop_footnote_columns_dbf(df: pandas.DataFrame) → pandas.DataFrame

Drop DBF footnote reference columns, which all end with _f.

source_table_primary_key(source_ferc1: SourceFerc1) → list[str]

Look up the pre-renaming source table primary key columns.

renamed_table_primary_key(source_ferc1: SourceFerc1) → list[str]

Look up the post-renaming primary key columns.

drop_unused_original_columns_dbf(df: pandas.DataFrame) → pandas.DataFrame

Remove residual DBF specific columns.

assign_record_id(df: pandas.DataFrame, source_ferc1: SourceFerc1) → pandas.DataFrame

Add a column identifying the original source record for each row.

It is often useful to be able to tell exactly which record in the FERC Form 1 database a given record within the PUDL database came from.

Within each FERC Form 1 DBF table, each record is supposed to be uniquely identified by the combination of: report_year, report_prd, utility_id_ferc1_dbf, spplmnt_num, row_number.

The FERC Form 1 XBRL tables do not have these supplement and row number columns, so we construct an id based on: report_year, utility_id_ferc1_xbrl, and the primary key columns of the XBRL table

Parameters:

• df – table to assign record_id to

• source_ferc1 – data source of raw ferc1 database.

Raises:

• ValueError – If any of the primary key columns are missing from the DataFrame being processed.

• ValueError – If there are any null values in the primary key columns.

• ValueError – If the resulting record_id column is non-unique.

assign_utility_id_ferc1(df: pandas.DataFrame, source_ferc1: SourceFerc1) → pandas.DataFrame

Assign the PUDL-assigned utility_id_ferc1 based on the native utility ID.

We need to replace the natively reported utility ID from each of the two FERC1 sources with a PUDL-assigned utilty. The mapping between the native ID’s and these PUDL-assigned ID’s can be accessed in the database tables utilities_dbf_ferc1 and utilities_xbrl_ferc1.

Parameters:

• df – the input table with the native utilty ID column.

• source_ferc1 – the

Returns:

an augemented version of the input df with a new column that replaces the natively reported utility ID with the PUDL-assigned utility ID.

reconcile_table_calculations(df: pandas.DataFrame, params: ReconcileTableCalculations | None = None)

Check how well a table’s calculated values match reported values.

class pudl.transform.ferc1.FuelFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

A table transformer specific to the fuel_ferc1 table.

The fuel_ferc1 table reports data about fuel consumed by large thermal power plants in the plants_steam_ferc1 table. Each record in the steam table is typically associated with several records in the fuel table, with each fuel record reporting data for a particular type of fuel consumed by that plant over the course of a year. The fuel table presents several challenges.

The type of fuel, which is part of the primary key for the table, is a freeform string with hundreds of different nonstandard values. These strings are categorized manually and converted to fuel_type_code_pudl. Some values cannot be categorized and are set to other. In other string categorizations we set the unidentifiable values to NA, but in this table the fuel type is part of the primary key and primary keys cannot contain NA values.

This simplified categorization occasionally results in records with duplicate primary keys. In those cases the records are aggregated into a single record if they have the same apparent physical units. If the fuel units are different, only the first record is retained.

Several columns have unspecified, inconsistent, fuel-type specific units of measure associated with them. In order for records to be comparable and aggregatable, we have to infer and standardize these units.

In the raw FERC Form 1 data there is a fuel_units column which describes the units of fuel delivered or consumed. Most commonly this is short tons for solid fuels (coal), thousands of cubic feet (Mcf) for gaseous fuels, and barrels (bbl) for liquid fuels. However, the fuel_units column is also a freeform string with hundreds of nonstandard values which we have to manually categorize, and many of the values do not map directly to the most commonly used units for fuel quantities. E.g. some solid fuel quantities are reported in pounds, or thousands of pounds, not tons; some liquid fuels are reported in gallons or thousands of gallons, not barrels; and some gaseous fuels are reported in cubic feet not thousands of cubic feet.

Two additional columns report fuel price per unit of heat content and fuel heat content per physical unit of fuel. The units of those columns are not explicitly reported, vary by fuel, and are inconsistent within individual fuel types.

We adopt standardized units and attempt to convert all reported values in the fuel table into those units. For physical fuel units we adopt those that are used by the EIA: short tons (tons) for solid fuels, barrels (bbl) for liquid fuels, and thousands of cubic feet (mcf) for gaseous fuels. For heat content per (physical) unit of fuel, we use millions of British thermal units (mmbtu). All fuel prices are converted to US dollars, while many are reported in cents.

Because the reported fuel price and heat content units are implicit, we have to infer them based on observed values. This is only possible because these quantities are ratios with well defined ranges of valid values. The common units that we observe and attempt to standardize include:

• coal: primarily BTU/pound, but also MMBTU/ton and MMBTU/pound.

• oil: primarily BTU/gallon.

• gas: reported in a mix of MMBTU/cubic foot, and MMBTU/thousand cubic feet.

table_id: TableIdFerc1

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Table specific transforms for fuel_ferc1.

Parameters:

df – Pre-processed, concatenated XBRL and DBF data.

Returns:

A single transformed table concatenating multiple years of cleaned data derived from the raw DBF and/or XBRL inputs.

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

Start with inherited method and do some fuel-specific processing.

We have to do most of the transformation before the DBF and XBRL data have been concatenated because the fuel type column is part of the primary key and it is extensively modified in the cleaning process.

process_xbrl(raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → pandas.DataFrame

Special pre-concat treatment of the fuel_ferc1 table.

We have to do most of the transformation before the DBF and XBRL data have been concatenated because the fuel type column is part of the primary key and it is extensively modified in the cleaning process. For the XBRL data, this means we can’t create a record ID until that fuel type value is clean. In addition, the categorization of fuel types results in a number of duplicate fuel records which need to be aggregated.

Parameters:

• raw_xbrl_instant – Freshly extracted XBRL instant fact table.

• raw_xbrl_duration – Freshly extracted XBRL duration fact table.

Returns:

Almost fully transformed XBRL data table, with instant and duration facts merged together.

standardize_physical_fuel_units(df: pandas.DataFrame) → pandas.DataFrame

Convert reported fuel quantities to standard units depending on fuel type.

Use the categorized fuel type and reported fuel units to convert all fuel quantities to the following standard units, depending on whether the fuel is a solid, liquid, or gas. When a single fuel reports its quantity in fundamentally different units, convert based on typical values. E.g. 19.85 MMBTU per ton of coal, 1.037 Mcf per MMBTU of natural gas, 7.46 barrels per ton of oil.

• solid fuels (coal and waste): short tons [ton]

• liquid fuels (oil): barrels [bbl]

• gaseous fuels (gas): thousands of cubic feet [mcf]

Columns to which these physical units apply:

• fuel_consumed_units (tons, bbl, mcf)

• fuel_cost_per_unit_burned (usd/ton, usd/bbl, usd/mcf)

• fuel_cost_per_unit_delivered (usd/ton, usd/bbl, usd/mcf)

One remaining challenge in this standardization is that nuclear fuel is reported in both mass of Uranium and fuel heat content, and it’s unclear if there’s any reasonable typical conversion between these units, since available heat content depends on the degree of U235 enrichement, the type of reactor, and whether the fuel is just Uranium, or a mix of Uranium and Plutonium from decommissioned nuclear weapons. See:

https://world-nuclear.org/information-library/facts-and-figures/heat-values-of-various-fuels.aspx

aggregate_duplicate_fuel_types_xbrl(fuel_xbrl: pandas.DataFrame) → pandas.DataFrame

Aggregate the fuel records having duplicate primary keys.

drop_total_rows(df: pandas.DataFrame) → pandas.DataFrame

Drop rows that represent plant totals rather than individual fuels.

This is an imperfect, heuristic process. The rows we identify as probably representing totals rather than individual fuels:

• have zero or null values in all of their numerical data columns

• have no identifiable fuel type

• have no identifiable fuel units

• DO report a value for MMBTU / MWh (heat rate)

In the case of the fuel_ferc1 table, we drop any row where all the data columns are null AND there’s a non-null value in the fuel_mmbtu_per_mwh column, as it typically indicates a “total” row for a plant. We also require a null value for the fuel_units and an “other” value for the fuel type.

drop_invalid_rows(df: pandas.DataFrame, params: pudl.transform.classes.InvalidRows | None = None) → pandas.DataFrame

Drop invalid rows from the fuel table.

This method both drops rows in which all required data columns are null (using the inherited parameterized method) and then also drops those rows we believe represent plant totals. See FuelFerc1TableTransformer.drop_total_rows().

class pudl.transform.ferc1.PlantsSteamFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for the plants_steam_ferc1 table.

table_id: TableIdFerc1

transform_main(df: pandas.DataFrame, transformed_fuel: pandas.DataFrame) → pandas.DataFrame

Perform table transformations for the plants_steam_ferc1 table.

Note that this method has a non-standard call signature, since the plants_steam_ferc1 table depends on the fuel_ferc1 table.

Parameters:

• df – The pre-processed steam plants table.

• transformed_fuel – The fully transformed fuel_ferc1 table. This is required because fuel consumption information is used to help link steam plant records together across years using plants_steam_assign_plant_ids()

transform(raw_dbf: pandas.DataFrame, raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame, transformed_fuel: pandas.DataFrame) → pandas.DataFrame

Redfine the transform method to accommodate the use of transformed_fuel.

This is duplicating code from the parent class, but is necessary because the steam table needs the fuel table for its transform. Is there a better way to do this that doesn’t require cutting and pasting the whole method just to stick the extra dataframe input into transform_main()?

class pudl.transform.ferc1.PlantsHydroFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

A table transformer specific to the plants_hydro_ferc1 table.

table_id: TableIdFerc1

transform_main(df)

Add bespoke removal of duplicate record after standard transform_main.

targeted_drop_duplicates(df)

Targeted removal of known duplicate record.

There are two records in 2019 with a utility_id_ferc1 of 200 and a plant_name_ferc1 of “marmet”. The records are nearly duplicates of eachother, except one have nulls in the capex columns. Surgically remove the record with the nulls.

class pudl.transform.ferc1.PlantsPumpedStorageFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for plants_pumped_storage_ferc1 table.

table_id: TableIdFerc1

class pudl.transform.ferc1.PurchasedPowerFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for purchased_power_ferc1 table.

This table has data about inter-utility power purchases into the PUDL DB. This includes how much electricty was purchased, how much it cost, and who it was purchased from. Unfortunately the field describing which other utility the power was being bought from is poorly standardized, making it difficult to correlate with other data. It will need to be categorized by hand or with some fuzzy matching eventually.

table_id: TableIdFerc1

class pudl.transform.ferc1.PlantInServiceFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

A transformer for the plant_in_service_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

process_xbrl_metadata(xbrl_metadata_converted: pandas.DataFrame, xbrl_calculations: pandas.DataFrame) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

We fill in some gaps in the metadata, e.g. for FERC accounts that have been split across multiple rows, or combined without being calculated. We also need to rename the XBRL metadata categories to conform to the same naming convention that we are using in the data itself (since FERC doesn’t quite follow their own naming conventions…). We use the same rename dictionary, but as an argument to pd.Series.replace() instead of pd.DataFrame.rename().

deduplicate_xbrl_factoid_xbrl_metadata(tbl_meta: pandas.DataFrame) → pandas.DataFrame

De-duplicate the XBLR metadata.

We deduplicate the metadata on the basis of the xbrl_factoid name. This table in particular has multiple wide_to_tidy value_types because there are multiple dollar columns embedded (it has both the standard start/end balances as well as modifcations like transfers/retirements). In the XBRL metadata, each xbrl_fact has its own set of metadata and possibly its own set of calculations. Which means that one xbrl_factoid for this table natively could have multiple calculations or other metadata.

For merging, we need the metadata to have one field per xbrl_factoid. Because we normally only use the start/end balance in calculations, when there are duplicate renamed xbrl_factoid s in our processed metadata, we are going to prefer the one that refers to the start/end balances. In an ideal world, we would be able to access this metadata based on both the xbrl_factoid and any column from value_types but that would require a larger change in architecture.

apply_sign_conventions(df) → pandas.DataFrame

Adjust rows and column sign conventsion to enable aggregation by summing.

Columns have uniform sign conventions, which we have manually inferred from the original metadata. This can and probably should be done programmatically in the future. If not, we’ll probably want to store the column_weights as a parameter rather than hard-coding it in here.

targeted_drop_duplicates_dbf(df: pandas.DataFrame) → pandas.DataFrame

Drop bad duplicate records from a specific utility in 2018.

This is a very specific fix, meant to get rid of a particular observed set of duplicate records: FERC Respondent ID 187 in 2018 has two sets of plant in service records, one of which contains a bunch of null data.

This method is part of the DBF processing because we want to be able to hard-code a specific value of utility_id_ferc1_dbf and those IDs are no longer available later in the process. I think.

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

Drop targeted duplicates in the DBF data so we can use FERC respondent ID.

transform_main(df: pandas.DataFrame) → pandas.DataFrame

The main table-specific transformations, affecting contents not structure.

Annotates and alters data based on information from the XBRL taxonomy metadata. Also assigns utility type for use in table explosions. Make all electric_plant_sold balances positive.

class pudl.transform.ferc1.PlantsSmallFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

A table transformer specific to the plants_small_ferc1 table.

table_id: TableIdFerc1

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Table specific transforms for plants_small_ferc1.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

A single transformed table concatenating multiple years of cleaned data derived from the raw DBF and/or XBRL inputs.

extract_ferc1_license(df: pandas.DataFrame) → pandas.DataFrame

Extract FERC license number from plant_name_ferc1.

Many FERC license numbers are embedded in the plant_name_ferc1 column, but not all numbers in the plant_name_ferc1 column are FERC licenses. Some are dates, dollar amounts, page numbers, or numbers of wind turbines. This function extracts valid FERC license numbers and puts them in a new column called license_id_ferc1.

Potential FERC license numbers are valid when:

• Two or more integers were found.

• The found integers were accompanied by key phrases such as: ["license", "no.", "ferc", "project"].

• The accompanying name does not contain phrases such as: ["page", "pg", "$",  "wind", "units"].

• The found integers don’t fall don’t fall within the range of a valid year, defined as: 1900-2050.

• The plant record is categorized as hydro or not categorized via the plant_type and fuel_type columns.

This function also fills other fuel types with hydro for all plants with valid FERC licenses because only hydro plants have FERC licenses.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

The same input DataFrame but with a new column called license_id_ferc1 that contains FERC 1 license infromation extracted from plant_name_ferc1.

_find_possible_header_or_note_rows(df: pandas.DataFrame) → pandas.DataFrame

Find and label rows that might be headers or notes.

Called by the coordinating function label_row_types().

This function creates a column called possible_header_or_note that is either True or False depending on whether a group of columns are all NA. Rows labeled as True will be further scrutinized in the _label_header_rows() and _label_note_rows() functions to determine whether they are actually headers or notes.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

The same input DataFrame but with a new column called possible_header_or_note that flags rows that might contain useful header or note information.

_find_note_clumps(group: pandas.core.groupby.DataFrameGroupBy) → tuple[pandas.core.groupby.DataFrameGroupBy, pandas.DataFrame]

Find groups of rows likely to be notes.

Once the _find_possible_header_or_note_rows() function identifies rows that are either headers or notes, we must deterine which one they are. As described in the _label_note_rows() function, notes rows are usually adjecent rows with no content.

This function itentifies instances of two or more adjecent rows where possible_header_or_note = True. It takes individual utility-year groups as a parameter as opposed to the entire dataset because adjecent rows are only meaningful if they are from the same reporting entity in the same year. If we were to run this on the whole dataframe, we would see “note clumps” that are actually notes from the end of one utility’s report and headers from the beginning of another. For this reason, we run this function from within the _label_note_rows_group() function.

The output of this function is not a modified version of the original utility-year group, rather, it is a DataFrame containing information about the nature of the possible_header_or_note = True rows that is used to determine if that row is a note or not. It also returns the original utility-year-group as groupby objects seperated by each time possible_header_or_note changes from True to False or vice versa.

If you pass in the following df:

plant_name_ferc1	possible_header_or_note
HYDRO:	True
rainbow falls (b)	False
cadyville (a)	False
keuka (c)	False
project #2738	True
project #2835	True
project #2852	True

You will get the following output (in addition to the groupby objects for each clump):

header_or_note	rows_per_clump
True	1
False	3
True	3

This shows each clump of adjecent records where possible_header_or_note is True or False and how many records are in each clump.

Params:

group: A utility-year grouping of the concatenated FERC XBRL and DBF tables.

This table must have been run through the _find_possible_header_or_note_rows() function and contain the column possible_header_or_note.

Returns:

A tuple containing groupby objects for each of the note and non-note clumps and a DataFrame indicating the number of rows in each note or non-note clump.

_label_header_rows(df: pandas.DataFrame) → pandas.DataFrame

Label header rows by adding header to row_type column.

Called by the coordinating function label_row_types().

Once possible header or notes rows have been identified via the _find_possible_header_or_note_rows() function, this function sorts out which ones are headers. It does this by identifying a list of strings that, when found in the plant_name_ferc1 column, indicate that the row is or is not a header.

Sometimes this function identifies a header that is acutally a note. For this reason, it’s important that the function be called before _label_note_rows() so that the bad header values get overridden by the note designation.

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through the _find_possible_header_or_note_rows() function and contains the column possible_header_or_note.

Returns:

The same input DataFrame but with likely headers rows containing the string header in the row_type column.

_label_note_rows_group(util_year_group: pandas.core.groupby.DataFrameGroupBy) → pandas.core.groupby.DataFrameGroupBy

Label note rows by adding note to row_type column.

Called within the wraper function _label_note_rows()

This function breaks the data down by reporting unit (utility and year) and determines whether a possible_header_note = True row is a note based on two criteria:

• Clumps of 2 or more adjecent rows where possible_header_or_note is True.

• Instances where the last row in a utility-year group has possible_header_or_note as True.

There are a couple of important exceptions that this function also addresses. Utilities often have multiple headers in a single utility-year grouping. You might see something like: pd.Series([header, plant1, plant2, note, header, plant3, plant4]). In this case, a note clump is actually comprised of a note followed by a header. This function will not override the header as a note. Unfortunately, there is always the possability that a header row is followed by a plant that had no values reported. This would look like, and therefore be categorized as a note clump. I haven’t built a work around, but hopefully there aren’t very many of these.

Params:

util_year_group: A groupby object that contains a single year and utility.

Returns:

The same input but with likely note rows containing the string note in the row_type column.

_label_note_rows(df: pandas.DataFrame) → pandas.DataFrame

Wrapper for _label_note_rows_group().

The small plants table has lots of note rows that contain useful information. Unfortunately, the notes are in their own row rather than their own column! This means that useful information pertaining to plant rows is floating around as a junk row with no other information except the note in the plant_name_ferc1 field. Luckily, the data are reported just like they would be on paper. I.e., The headers are at the top, and the notes are at the bottom. See the table in label_row_types() for more detail. This function labels note rows.

Note rows are determined by row location within a given report, so we must break the data into reporting units (utility and year) and then apply note-finding methodology defined in _label_note_rows_group() to each group.

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through the _find_possible_header_or_note_rows() function and contains the column possible_header_or_note.

Returns:

The same input DataFrame but with likely note rows containing the string note in the row_type column.

_label_total_rows(df: pandas.DataFrame) → pandas.DataFrame

Label total rows by adding total to row_type column.

Called within the wraper function _label_note_rows()

For the most part, when plant_name_ferc1 contains the string total, the values therein are duplicates of what is already reported, i.e.: a total value. However, there are some cases where that’s not true. For example, the phrase amounts are for the total appears when chunks of plants (usually but not always wind) are reported together. It’s a total, but it’s not double counting which is the reason for the total flag.

Similar to _label_header_rows(), it’s important that this be called before _label_note_rows() in label_row_types() so that not clumps can override certain non-totals that are mistakenly labeled as such.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

The same input DataFrame but with likely total rows containing the string total in the row_type column.

label_row_types(df: pandas.DataFrame) → pandas.DataFrame

Coordinate labeling of row_types as headers, notes, or totals.

The small plants table is more like a digitized PDF than an actual data table. The rows contain all sorts of information in addition to what the columns might suggest. For instance, there are header rows, note rows, and total rows that contain useful information, but cause confusion in their current state, mixed in with the rest of the data.

Here’s an example of what you might find in the small plants table:

plant_name_ferc1	plant_type	capacity_mw
HYDRO:	NA	NA
rainbow falls (b)	NA	30
cadyville (a)	NA	100
keuka (c)	NA	80
total plants	NA	310
project #2738	NA	NA
project #2835	NA	NA
project #2852	NA	NA

Notice how misleading it is to have all this infomration in one column. The goal of this function is to coordinate labeling functions so that we can identify which rows contain specific plant information and which rows are headers, notes, or totals.

Once labeled, other functions can either remove rows that might cause double counting, extract useful plant or fuel type information from headers, and extract useful context or license id information from notes.

Coordinates _label_header_rows(), _label_total_rows(), _label_note_rows().

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through the _find_possible_header_or_note_rows() function and contains the column possible_header_or_note.

Returns:

The same input DataFrame but with a column called row_type containg the strings header, note, total, or NA to indicate what type of row it is.

prep_header_fuel_and_plant_types(df: pandas.DataFrame, show_unmapped_headers=False) → pandas.DataFrame

Forward fill header rows to prep for fuel and plant type extraction.

The headers we’ve identified in _label_header_rows() can be used to supplement the values in the plant_type and fuel_type columns.

This function groups the data by utility, year, and header; extracts the header into a new column; and forward fills the headers so that each record in the header group is associated with that header. Because the headers map to different fuel types and plant types (ex: solar pv maps to fuel type solar and plant type photovoltaic), the new forward-filled header column is duplicated and called fuel_type_from_header and plant_type_from_header. In map_header_fuel_and_plant_types(), these columns will be mapped to their respective fuel and plant types, used to fill in blank values in the plant_type and fuel_type, and then eventually removed.

Why separate the prep step from the map step?

We trust the values originally reported in the fuel_type and plant_type columns more than the extracted and forward filled header values, so we only want to replace fuel_type and plant_type values that are labeled as pd.NA or other. The values reported to those columns are extremely messy and must be cleaned via pudl.transform.classes.categorize_strings() in order for us to know which are truely pd.NA or other. Because we also use pudl.transform.classes.categorize_strings() to map the headers to fuel and plant types, it makes sense to clean all four columns at once and then combine them.

Here’s a look at what this function does. It starts with the following table:

plant_name_ferc1	plant_type	fuel_type	row_type
HYDRO:	NA	NA	header
rainbow falls (b)	NA	NA	NA
cadyville (a)	NA	NA	NA
keuka (c)	NA	NA	NA
Wind Turbines:	NA	NA	header
sunny grove	NA	NA	NA
green park wind	NA	wind	NA

And ends with this:

plant_name_ferc1	plant _type	fuel _type	plant_type _from_header	fuel_type _from_header
HYDRO:	NA	NA	HYDRO:	HYDRO:
rainbow falls (b)	NA	NA	HYDRO:	HYDRO:
cadyville (a)	NA	NA	HYDRO:	HYDRO:
keuka (c)	NA	NA	HYDRO:	HYDRO:
Wind Turbines:	NA	NA	Wind Turbines:	Wind Turbines:
sunny grove	NA	NA	Wind Turbines:	Wind Turbines:
green park wind	NA	wind	Wind Turbines:	Wind Turbines:

NOTE: If a utility’s plant_name_ferc1 values look like this: ["STEAM", "coal_plant1", "coal_plant2", "wind_turbine1"], then this algorythem will think that last wind turbine is a steam plant. Luckily, when a utility embeds headers in the data it usually includes them for all plant types: ["STEAM", "coal_plant1", "coal_plant2", "WIND", "wind_turbine"].

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through _label_row_type() and contains the columns row_type.

Returns:

The same input DataFrame but with new columns plant_type_from_header and fuel_type_from_header that forward fill the values in the header rows by utility, year, and header group.

map_header_fuel_and_plant_types(df: pandas.DataFrame) → pandas.DataFrame

Fill pd.NA and other plant and fuel types with cleaned headers.

prep_header_fuel_and_plant_types() extracted and forward filled the header values; pudl.transform.params.categorize_strings() cleaned them according to both the fuel and plant type parameters. This function combines the fuel_type_from_header with fuel_type and plant_type_from_header with plant_type when the reported, cleaned values are pd.NA or other.

To understand more about why these steps are necessary read the docstrings for prep_header_fuel_and_plant_types().

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through prep_header_fuel_and_plant_types() and contains the columns fuel_type_from_header and plant_type_from_header.

Returns:

The same input DataFrame but with rows with pd.NA or other in the fuel_type and plant_type columns filled in with the respective values from fuel_type_from_header and plant_type_from_header when available. fuel_type_from_header and plant_type_from_header columns removed.

map_plant_name_fuel_types(df: pandas.DataFrame) → pandas.DataFrame

Suppliment fuel_type with information in plant_name_ferc1.

Sometimes fuel type is embedded in a plant name (not just headers). In this case we can identify that what that fuel is from the name and fill in empty fuel_type values. Right now, this only works for hydro plants because the rest are complicated and have a slew of exceptions. This could probably be applied to the plant_type column in the future too.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

The same input DataFrame but with rows with other in the fuel_type column filled in notable fuel types extracted from the the plant_name_ferc1 column.

associate_notes_with_values(df: pandas.DataFrame) → pandas.DataFrame

Use footnote indicators to map notes and FERC licenses to plant rows.

There are many utilities that report a bunch of mostly empty note rows at the bottom of their yearly entry. These notes often pertain to specific plant rows above. Sometimes the notes and their respective plant rows are linked by a common footnote indicator such as (a) or (1) etc.

This function takes this:

plant_name_ferc1	row_type	license_id_ferc1
HYDRO:	header	NA
rainbow falls (b)	NA	NA
cadyville (a)	NA	NA
keuka (c)	NA	NA
total plants	total	NA
project #2738	note	2738
project #2835	note	2738
project #2852	note	2738

Finds the note rows with footnote indicators, maps the content from the note row into a new note column that’s associated with the value row, and maps any FERC license extracted from this note column to the license_id_ferc1 column in the value row.

plant_name_ferc1	row_type	notes	license_id_ferc1
HYDRO:	header	NA	NA
rainbow falls (b)	NA	project #2835	2835
cadyville (a)	NA	project #2738	2738
keuka (c)	NA	project #2852	2752
total plants	total	NA	NA
project #2738	note	NA	2738
project #2835	note	NA	2835
project #2852	note	NA	2752

(Header and note rows are removed later).

NOTE: Note rows that don’t have a footnote indicator or note rows with a footnote indicator that don’t have a cooresponding plant row with the same indicator are not captured. They will ultimately get removed and their content will not be preserved.

Params:

df: Pre-processed, concatenated XBRL and DBF data that has been run through label_row_types() and contains the column row_type.

Returns:

The same input DataFrame but with a column called notes that contains notes, reported below, in the same row as the plant values they pertain to. Also, any further additions to the license_id_ferc1 field as extracted from these newly associated notes.

spot_fix_rows(df: pandas.DataFrame) → pandas.DataFrame

Fix one-off row errors.

In 2004, utility_id_ferc1 251 reports clumps of units together. Each unit clump looks something like this: intrepid wind farm (107 units @ 1.5 mw each) and is followed by a row that looks like this: (amounts are for the total of all 107 units). For the most part, these rows are useless note rows. However, there is one instance where important values are reported in this note row rather than in the actual plant row above.

There are probably plenty of other spot fixes one could add here.

Params:

df: Pre-processed, concatenated XBRL and DBF data.

Returns:

The same input DataFrame but with some spot fixes corrected.

class pudl.transform.ferc1.TransmissionStatisticsFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

A table transformer for the transmission_statistics_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

class pudl.transform.ferc1.ElectricEnergySourcesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for electric_energy_sources_ferc1 table.

The raw DBF and XBRL table will be split up into two tables. This transformer generates the sources of electricity for utilities, dropping the information about dispositions. For XBRL, this is a duration-only table. Right now we are merging in the metadata but not actually keeping anything from it. We are also not yet doing anything with the sign.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Perform default xbrl metadata processing plus adding 1 new xbrl_factoid.

Note: we should probably parameterize this and add it into the standard process_xbrl_metadata().

class pudl.transform.ferc1.ElectricEnergyDispositionsFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for electric_energy_dispositions_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

class pudl.transform.ferc1.UtilityPlantSummaryFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for utility_plant_summary_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Do the default metadata processing plus add a new factoid.

The new factoid cooresponds to the aggregated factoid in aggregated_xbrl_factoids().

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Default transforming, plus spot fixing and building aggregate xbrl_factoid.

aggregated_xbrl_factoids(df: pandas.DataFrame) → pandas.DataFrame

Aggregate xbrl_factoids records for linking to plant_in_service_ferc1.

This table has two xbrl_factoid which can be linked via calcuations to one xbrl_factoid in the plant_in_service_ferc1. Doing this 2:1 linkage would be fine in theory. But the plant_in_service_ferc1 is in most senses the table with the more details and of our desire to build tree-link relationships between factoids, we need to build a new factoid to link in a 1:1 manner between this table and the plant_in_service_ferc1.

We’ll also add this factoid into the metadata via process_xbrl_metadata() and add the linking calculation via apply_xbrl_calculation_fixes().

spot_fix_bad_signs(df: pandas.DataFrame) → pandas.DataFrame

Spot fix depreciation_utility_plant_in_service records with bad signs.

class pudl.transform.ferc1.BalanceSheetLiabilitiesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for balance_sheet_liabilities_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Duplicate data that appears in multiple distinct calculations.

There is a one case in which exactly the same data values are referenced in multiple calculations which can’t be resolved by choosing one of the referenced values as the canonical location for that data. In order to preserve all of the calculation structure, we need to duplicate those records in the data, the metadata, and the calculation specifications. Here we duplicate the data and associated it with newly defined facts, which we will also add to the metadata and calculations.

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Perform default xbrl metadata processing plus adding 2 new xbrl_factoids.

We add two new factoids which are defined (by PUDL) only for the DBF data, and also duplicate and redefine several factoids which are referenced in multiple calculations and need to be distinguishable from each other.

Note: we should probably parameterize this and add it into the standard process_xbrl_metadata().

class pudl.transform.ferc1.BalanceSheetAssetsFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for balance_sheet_assets_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Duplicate data that appears in multiple distinct calculations.

There is a one case in which exactly the same data values are referenced in multiple calculations which can’t be resolved by choosing one of the referenced values as the canonical location for that data. In order to preserve all of the calculation structure, we need to duplicate those records in the data, the metadata, and the calculation specifications. Here we duplicate the data and associated it with newly defined facts, which we will also add to the metadata and calculations.

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Default xbrl metadata processing plus some error correction.

We add two new factoids which are defined (by PUDL) only for the DBF data, and also duplicate and redefine several factoids which are referenced in multiple calculations and need to be distinguishable from each other.

Note: we should probably parameterize this and add it into the standard process_xbrl_metadata().

class pudl.transform.ferc1.IncomeStatementFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for the income_statement_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Perform default xbrl metadata processing plus adding a new xbrl_factoid.

Note: we should probably parameterize this and add it into the standard process_xbrl_metadata().

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

Drop incorrect row numbers from f1_incm_stmnt_2 before standard processing.

In 2003, two rows were added to the f1_income_stmnt dbf table, which bumped the starting row_number of f1_incm_stmnt_2 from 25 to 27. A small handfull of respondents seem to have not gotten the memo about this this in 2003 and have information on these row numbers that shouldn’t exist at all for this table.

This step necessitates the ability to know which source table each record actually comes from, which required adding a column (sched_table_name) in the extract step before these two dbf input tables were concatenated.

Right now we are just dropping these bad row numbers. Should we actually be bumping the whole respondent’s row numbers - assuming they reported incorrectly for the whole table? See: https://github.com/catalyst-cooperative/pudl/issues/471

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Drop duplicate records from f1_income_stmnt.

Because net_utility_operating_income is reported on both page 1 and 2 of the form, it ends up introducing a bunch of duplicated records, so we need to drop one of them. Since the value is used in the calculations that are part of the second page, we’ll drop it from the first page.

class pudl.transform.ferc1.RetainedEarningsFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for retained_earnings_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

Beyond the standard Ferc1AbstractTableTransformer.process_xbrl_metadata() processing, add FERC account values for a few known values.

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

Preform generic process_dbf(), plus deal with duplicates.

Along with the standard processing in Ferc1AbstractTableTransformer.process_dbf(), this method runs: * targeted_drop_duplicates_dbf() * reconcile_double_year_earnings_types_dbf()

transform_main(df)

Add _previous_year factoids after standard transform_main.

Add _previous_year factoids for unappropriated_retained_earnings and unappropriated_undistributed_subsidiary_earnings after standard transform_main. This should only affect XBRL data, but we do it after merging to enable access to DBF data to fill this in as well.

targeted_drop_duplicates_dbf(df: pandas.DataFrame) → pandas.DataFrame

Drop duplicates with truly duplicate data.

There are instances of utilities that reported multiple values for several earnings types for a specific year (utility_id_ferc1 68 in 1998 & utility_id_ferc1 296 in 2015). We are taking the largest value reported and dropping the rest. There very well could be a better strategey here, but there are only 25 records that have this problem, so we’ve going with this.

reconcile_double_year_earnings_types_dbf(df: pandas.DataFrame) → pandas.DataFrame

Reconcile current and past year data reported in 1 report_year.

The DBF table includes two different earnings types that have: “Begining of Period” and “End of Period” rows. But the table has both an amount column that corresponds to a balance and a starting balance column. For these two earnings types, this means that there is in effect two years of data in this table for each report year: a starting and ending balance for the pervious year and a starting and ending balance for the current year. The ending balance for the previous year should be the same as the starting balance for the current year.

We need to keep both pieces of data in order to calculate ending_balances, so we want to check these assumptions, extract as much information from these two years of data, and keep both records for each of these two earnings types for each utility.

Raises:

AssertionError – There are a very small number of instances in which the ending balance from the previous year does not match the starting balance from the current year. The % of these non-matching instances should be less than 2% of the records with these date duplicative earnings types.

add_previous_year_factoid(df: pandas.DataFrame) → pandas.DataFrame

Add previous_year factoids to XBRL data from prior year’s DBF data.

deduplicate_xbrl_factoid_xbrl_metadata(tbl_meta) → pandas.DataFrame

Deduplicate the xbrl_metadata based on the xbrl_factoid.

The metadata relating to dollar_value column generally had the same name as the renamed xbrl_factoid. we’ll double check that we a) didn’t remove too many factoid’s by doing this AND that we have a fully deduped output below. In an ideal world, we would have multiple pieces of metadata information (like calucations and ferc account #’s), for every single wide_to_tidy() value column.

Note: This is almost the same as the method for electric_operating_revenues_ferc1. If we wanted to lean into this version of deduplication more generally this might be a fine way start to an abstraction, but ideally we wouldn’t need to dedupe this at all and instead enable metadata for every value column from wide_to_tidy().

class pudl.transform.ferc1.DepreciationAmortizationSummaryFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for depreciation_amortization_summary_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

process_xbrl_metadata(xbrl_metadata_converted: pandas.DataFrame, xbrl_calculations: pandas.DataFrame) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

Beyond the standard Ferc1AbstractTableTransformer.process_xbrl_metadata() processing, add FERC account values for a few known values.

transform_main(df)

After standard transform_main, assign utility type as electric.

class pudl.transform.ferc1.ElectricPlantDepreciationChangesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for electric_plant_depreciation_changes_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

Warning: The calculations in this table are currently being corrected using reconcile_table_calculations(), but they still contain high rates of error. This function replaces the name of the single balance column reported in the XBRL Instant table with starting_balance / ending_balance. We pull those two values into their own separate labeled rows, each of which should get the metadata from the original column. We do this pre-processing before we call the main function in order for the calculation fixes and renaming to work as expected.

process_dbf(raw_df: pandas.DataFrame) → pandas.DataFrame

Accumulated Depreciation table specific DBF cleaning operations.

The XBRL reports a utility_type which is always electric in this table, but which may be necessary for differentiating between different values when this data is combined with other tables. The DBF data doesn’t report this value so we are adding it here for consistency across the two data sources.

Also rename the ending_balance_accounts to ending_balance

process_instant_xbrl(df: pandas.DataFrame) → pandas.DataFrame

Pre-processing required to make the instant and duration tables compatible.

This table has a rename that needs to take place in an unusual spot – after the starting / ending balances have been usntacked, but before the instant & duration tables are merged. This method just reversed the order in which these operations happen, comapared to the inherited method.

class pudl.transform.ferc1.ElectricPlantDepreciationFunctionalFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer for electric_plant_depreciation_functional_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Create a metadata table with the one factoid we’ve assigned to this table.

Instead of adding facts to the metdata like a lot of the other table-specific convert_xbrl_metadata_json_to_df(), this method creates a metadata table with one singular xbrl_factoid. We assign that factoid to the table in transform_main().

raw_xbrl_factoid_to_pudl_name(col_name_xbrl: str) → str

Return the one fact name for this table.

We’ve artificially assigned this table to have one xbrl_factoid during transform_main(). Because this table only has one value for its xbrl_factoid column, all col_name_xbrl should be converted to “accumulated_depreciation”.

process_dbf(raw_df: pandas.DataFrame) → pandas.DataFrame

Accumulated Depreciation table specific DBF cleaning operations.

The XBRL reports a utility_type which is always electric in this table, but which may be necessary for differentiating between different values when this data is combined with other tables. The DBF data doesn’t report this value so we are adding it here for consistency across the two data sources.

process_instant_xbrl(df: pandas.DataFrame) → pandas.DataFrame

Pre-processing required to make the instant and duration tables compatible.

This table has a rename that needs to take place in an unusual spot – after the starting / ending balances have been usntacked, but before the instant & duration tables are merged. This method reverses the order in which these operations happen comapared to the inherited method. We also want to strip the accumulated_depreciation that appears on every plant functional class.

transform_main(df: pandas.DataFrame) → pandas.DataFrame

Add depreciation_type then run default transform_main().

We are adding depreciation_type as the xbrl_factoid column for this table with one value (“accumulated_depreciation”) across the whole table. This table has multiple “dimension” columns such as utility_type and plant_function which differentiate what slice of a utility’s assets each record pertains to. We added this new column as the xbrl_factoid of the table instead of using one of the dimensions of the table so that the table can conform to the same patern of treatment for these dimension columns.

class pudl.transform.ferc1.ElectricOperatingExpensesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for electric_operating_expenses_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

targeted_drop_duplicates_dbf(raw_df: pandas.DataFrame) → pandas.DataFrame

Drop incorrect duplicate from 2002.

In 2002, utility_id_ferc1_dbf 96 reported two values for administrative_and_general_operation_expense. I found the correct value by looking at the prev_yr_amt value in 2003. This removes the incorrect row.

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Default XBRL metadata processing and add a DBF-only xblr factoid.

Note: we should probably parameterize this and add it into the standard process_xbrl_metadata().

process_dbf(raw_dbf: pandas.DataFrame) → pandas.DataFrame

Process DBF but drop a bad row that is flagged by drop_duplicates.

transform_main(df)

After standard transform_main, assign utility type as electric.

class pudl.transform.ferc1.ElectricOperatingRevenuesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for electric_operating_revenues_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

deduplicate_xbrl_factoid_xbrl_metadata(tbl_meta: pandas.DataFrame) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

Employ the standard process for processing metadata. Then remove duplication on the basis of the xbrl_factoid. This table used wide_to_tidy() with three seperate value columns. Which results in one xbrl_factoid referencing three seperate data columns. This method grabs only one piece of metadata for each renamed xbrl_factoid, preferring the calculated value or the factoid referencing the dollar columns.

In an ideal world, we would have multiple pieces of metadata information (like calucations and ferc account #’s), for every single wide_to_tidy() value column. We would probably want to employ that across the board - adding suffixes or something like that to stack the metadata in a similar fashion that we stack the data.

transform_main(df)

Add duplicate removal after standard transform_main & assign utility type.

targeted_drop_duplicates(df)

Drop one duplicate records from 2011, utility_id_ferc1 295.

class pudl.transform.ferc1.CashFlowFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transform class for cash_flow_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

process_instant_xbrl(df: pandas.DataFrame) → pandas.DataFrame

Pre-processing required to make the instant and duration tables compatible.

This table has a rename that needs to take place in an unusual spot – after the starting / ending balances have been usntacked, but before the instant & duration tables are merged. This method just reversed the order in which these operations happen, comapared to the inherited method.

transform_main(df)

Add duplicate removal and validation after standard transform_main.

targeted_drop_duplicates(df)

Drop one duplicate record from 2020, utility_id_ferc1 2037.

Note: This step could be avoided if we employed a drop_invalid_rows() transform step with required_valid_cols = ["amount"]

validate_start_end_balance(df)

Validate of start balance + net = end balance.

Add a quick check to ensure the vast majority of the ending balances are calculable from the net change + the starting balance = the ending balance.

convert_xbrl_metadata_json_to_df(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]]) → pandas.DataFrame

Transform the metadata to reflect the transformed data.

Replace the name of the balance column reported in the XBRL Instant table with starting_balance / ending_balance since we pull those two values into their own separate labeled rows, each of which should get the original metadata for the Instant column.

class pudl.transform.ferc1.ElectricitySalesByRateScheduleFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transform class for electricity_sales_by_rate_schedule_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids: bool = False

add_axis_to_total_table_rows(df: pandas.DataFrame)

Add total to the axis column for rows from the total table.

Because we’re adding the sales_of_electricity_by_rate_schedules_account_totals_304 table into the mix, we have a bunch of total values that get mixed in with all the _billed columns from the individual tables. If left alone, these totals aren’t labeled in any way becuse they don’t have the same _axis columns explaining what each of the values are. In order to distinguish them from the rest of the sub-total data we use this function to create an _axis value for them noting that they are totals.

It’s worth noting that there are also some total values in there already. Those would be hard to clean. The idea is that if you want the actual totals, don’t try and sum the sub-components, look at the actual labeled total rows.

This function relies on the sched_table_name column, so it must be called before that gets dropped.

Parameters:

df – The sales table with a sched_table_name column.

process_xbrl(raw_xbrl_instant: pandas.DataFrame, raw_xbrl_duration: pandas.DataFrame) → pandas.DataFrame

Rename columns before running wide_to_tidy.

class pudl.transform.ferc1.OtherRegulatoryLiabilitiesFerc1TableTransformer(xbrl_metadata_json: dict[Literal[instant, duration], list[dict[str, Any]]] | None = None, params: pudl.transform.classes.TableTransformParams | None = None, cache_dfs: bool = False, clear_cached_dfs: bool = True)

Bases: Ferc1AbstractTableTransformer

Transformer class for other_regulatory_liabilities_ferc1 table.

table_id: TableIdFerc1

has_unique_record_ids = False

pudl.transform.ferc1.FERC1_TFR_CLASSES: collections.abc.Mapping[str, type[Ferc1AbstractTableTransformer]]

pudl.transform.ferc1.ferc1_transform_asset_factory(table_name: str, tfr_class: Ferc1AbstractTableTransformer, io_manager_key: str = 'pudl_sqlite_io_manager', convert_dtypes: bool = True, generic: bool = False) → dagster.AssetsDefinition

Create an asset that pulls in raw ferc Form 1 assets and applies transformations.

This is a convenient way to create assets for tables that only depend on raw dbf, raw xbrl instant and duration tables and xbrl metadata. For tables with additional upstream dependencies, create a stand alone asset using an asset decorator. See the plants_steam_ferc1 asset.

Parameters:

• table_name – The name of the table to create an asset for.

• tfr_class – A transformer class cooresponding to the table_name.

• io_manager_key – the dagster io_manager key to use. None defaults to the fs_io_manager.

• convert_dtypes – convert dtypes of transformed dataframes.

• generic – If using GenericPlantFerc1TableTransformer pass table_id to constructor.

Returns:

An asset for the clean table.

pudl.transform.ferc1.create_ferc1_transform_assets() → list[dagster.AssetsDefinition]

Create a list of transformed FERC Form 1 assets.

Returns:

A list of AssetsDefinitions where each asset is a clean ferc form 1 table.

pudl.transform.ferc1.ferc1_assets

pudl.transform.ferc1.plants_steam_ferc1(clean_xbrl_metadata_json: dict[str, dict[str, list[dict[str, Any]]]], raw_ferc1_dbf__f1_steam: pandas.DataFrame, raw_ferc1_xbrl__steam_electric_generating_plant_statistics_large_plants_402_duration: pandas.DataFrame, raw_ferc1_xbrl__steam_electric_generating_plant_statistics_large_plants_402_instant: pandas.DataFrame, fuel_ferc1: pandas.DataFrame) → pandas.DataFrame

Create the clean plants_steam_ferc1 table.

Parameters:

• clean_xbrl_metadata_json – XBRL metadata json for all tables.

• raw_ferc1_dbf__f1_steam – Raw f1_steam table.

• raw_ferc1_xbrl__steam_electric_generating_plant_statistics_large_plants_402_duration – raw XBRL duration table.

• raw_ferc1_xbrl__steam_electric_generating_plant_statistics_large_plants_402_instant – raw XBRL instant table.

• fuel_ferc1 – Transformed fuel_ferc1 table.

Returns:

Clean plants_steam_ferc1 table.

pudl.transform.ferc1.other_dimensions(table_names: list[str]) → list[str]

Get a list of the other dimension columns across all of the transformers.

pudl.transform.ferc1.table_to_xbrl_factoid_name() → dict[str, str]

Build a dictionary of table name (keys) to xbrl_factiod column name.

pudl.transform.ferc1.table_dimensions_ferc1(**kwargs) → pandas.DataFrame

Build a table of values of dimensions observed in the transformed data tables.

Compile a dataframe indicating what distinct values are observed in the data for each dimension column in association with each unique combination of table_name and xbrl_factoid. E.g. for all factoids found in the electric_plant_depreciation_functional_ferc1 table, the only value observed for utility_type is electric and the values observed for plant_status include: future, in_service, leased and total.

We need to include the xbrl_factoid column because these dimensions can differ based on the xbrl_factoid. So we first rename all of the columns which contain the xbrl_factoid using table_to_xbrl_factoid_name() rename dictionary. Then we concatenate all of the tables together and drop duplicates so we have unique instances of observed table_name and xbrl_factoid and the other dimension columns found in other_dimensions().

pudl.transform.ferc1.metadata_xbrl_ferc1(**kwargs) → pandas.DataFrame

Build a table of all of the tables’ XBRL metadata.

pudl.transform.ferc1.calculation_components_xbrl_ferc1(**kwargs) → pandas.DataFrame

Create calculation-compnent table from table-level metadata.

pudl.transform.ferc1.unexpected_total_components(calc_comps: pandas.DataFrame, dimensions: list[str]) → pandas.DataFrame

Find unexpected components in within-fact total calculations.

This doesn’t check anything about the calcs we get from the metadata, we are only looking at within-fact totals which we’ve added ourselves.

Finds calculation relationships where:

• child components that do not match with parent in non-total dimensions.

  • For example, if utility_type_parent is not “total”, then utility_type must be the same as utility_type_parent.

• child components, that share table_name/xbrl_factoid with their parent, that have “total” for any dimension - these should be represented by their child components

Parameters:

• calc_comps – calculation component join table

• dimensions – list of dimensions we resolved “total” values for

pudl.transform.ferc1.check_for_calc_components_duplicates(calc_components: pandas.DataFrame, table_names_known_dupes: list[str], idx: list[str]) → None

Check for duplicates calculation records.

We need to remove the electricity_sales_by_rate_schedule_ferc1 bc there are duplicate renamed factoids in that table (originally billed/unbilled).

pudl.transform.ferc1.make_calculation_dimensions_explicit(calculation_components: pandas.DataFrame, table_dimensions_ferc1: pandas.DataFrame, dimensions: list[str], parent: bool = False) → pandas.DataFrame

Fill in null dimensions w/ the values observed in table_dimensions_ferc1().

In the raw XBRL metadata’s calculations, there is an implicit assumption that calculated values are aggregated within categorical columns called Axes or dimensions, in addition to being grouped by date, utility, table, and fact. The dimensions and their values don’t need to be specified explicitly in the calculation components because the same calculation is assumed to apply in all cases.

We have extended this calculation system to allow independent calculations to be specified for different values within a given dimension. For example, the utility_plant_summary_ferc1 table contains records with a variety of different utility_type values (gas, electric, etc.). For many combinations of fact and utility_type, no more detailed information about the soruce of the data is available, but for some, and only in the case of electric utilities, much more detail can be found in the plant_in_service_ferc1 table. In order to use this additional information when it is available, we sometimes explicitly specify different calculations for different values of additional dimension columns.

This function uses the observed associations between table_name, xbrl_factoid and the other dimension columns compiled by table_dimensions_ferc1() to fill in missing (previously implied) dimension values in the calculation components table.

This is often a broadcast merge because many tables contain many values within these dimension columns, so it is expected that new calculation component table will have many more records than the input calculation components table.

Any dimension that was already specified in the calculation fixes will be left unchanged. If no value of a particular dimension has ever been observed in association with a given combination of table_name and xbrl_factoid it will remain null.

Parameters:

• calculation_components – a table of calculation component records which have had some manual calculation fixes applied.

• table_dimensions_ferc1 – table with all observed values of other_dimensions() for each table_name and xbrl_factoid

• dimensions – list of dimension columns to check.

• parent – boolean to indicate whether or not the dimensions to be added are the parental dimensions or the child dimensions.

pudl.transform.ferc1.assign_parent_dimensions(calc_components: pandas.DataFrame, table_dimensions: pandas.DataFrame, dimensions: list[str]) → pandas.DataFrame

Add dimensions to calculation parents.

We now add in parent-dimension values for all of the original calculation component records using the observed dimensions.

Parameters:

• calc_components – a table of calculation component records which have had some manual calculation fixes applied.

• table_dimensions – table with all observed values of other_dimensions() for each table_name and xbrl_factoid.

• dimensions – list of dimension columns to check.

pudl.transform.ferc1.infer_intra_factoid_totals(calc_components: pandas.DataFrame, meta_w_dims: pandas.DataFrame, table_dimensions: pandas.DataFrame, dimensions: list[str]) → pandas.DataFrame

Define dimension total calculations.

Some factoids are marked as a total along some dimension in the metadata, which means that they are the sum of all the non-total factoids along that dimension.

We match the parent factoids from the metadata to child factoids from the table_dimensions. We treat “total” as a wildcard value.

We exclude child factoids that are themselves totals, because that would result in a double-count.

Here are a few examples:

Imagine a factoid with the following dimensions & values:

• utility types: “total”, “gas”, “electric”;

• plant status: “total”, “in_service”, “future”

Then the following parents would match/not-match:

• parent: “total”, “in_service”

  • child: “gas”, “in_service” WOULD match.

  • child: “electric”, “in_service” WOULD match.

  • child: “electric”, “future” WOULD NOT match.

• parent: “total”, “total”

  • child: “gas”, “in_service” WOULD match.

  • child: “electric”, “future” WOULD match.

See the unit test in ferc1_test.py for more details.

To be able to define these within-dimension calculations we also add dimension columns to all of the parent factoids in the table.

Parameters:

• calc_components – a table of calculation component records which have had some manual calculation fixes applied. Passed through unmodified.

• meta_w_dims – metadata table with the dimensions.

• table_dimensions – table with all observed values of other_dimensions() for each table_name and xbrl_factoid.

• dimensions – list of dimension columns to check.

Returns:

An table associating calculation components with the parents they will be aggregated into. The components and the parents are each identified by table_name, xbrl_factoid, and columns defining the additional dimensions (utility_type, plant_status, plant_function). The parent columns have a _parent suffix.