Skip to main content

post_processing

Utilities for post-processing.

Classes​

ColumnRenamerPostProcessor​

class ColumnRenamerPostProcessor(column_mapping: dict[str, str]):

Renames columns in a DataFrame.

Initialize the column renamer.

Arguments

  • column_mapping: Mapping from old column names to new column names.

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process the predictions by renaming columns.

Arguments

  • predictions: The model output to process.

Returns Processed predictions with renamed columns.

JSONExtractToColumnsPostProcessor​

class JSONExtractToColumnsPostProcessor(    source_column: str,    extractions: list[dict[str, Any]] | None = None,    drop_source: bool = True,):

Extracts fields from a JSON dict column into new DataFrame columns.

Uses dot-notation paths with support for integer array indexing.

Extraction configs are provided as a list of dicts, each with:

  • target_column – name of the new DataFrame column to create
  • source_path – dot-notation path to extract a value from the JSON dict (e.g. "exam_center_slice_index.0"). Supports integer indices for list access.

Initialise the JSON extract-to-columns postprocessor.

Arguments

  • source_column: Name of the column containing the parsed JSON dict.
  • extractions: List of extraction configs. Each must have target_column and source_path.
  • drop_source: Whether to drop the source column after extraction.

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Extract fields from JSON dicts into new columns.

JSONFieldRestructuringPostProcessor​

class JSONFieldRestructuringPostProcessor(    column_patterns: list[str],    field_mappings: list[dict[str, str]],    keep_original: bool = False,):

Restructures JSON data by moving fields between different levels.

This is done for example to move fields from one json level to another as for example from {"key1": {"key2": "value", "key3": "val"}} to {"key1": {"key2": "value"}, "key3": "val"} and vice-versa.

Initialize the JSON field restructurer.

Arguments

  • column_patterns: List of regex patterns matching columns to process
  • field_mappings: List of mappings, each with:
    • 'source_path': JSON path to extract (using dot notation, e.g., 'key1.key2' or 'key2')
    • 'target_path': Path to place the value (dot notation, e.g., ''key1.key2' or 'key2'')
  • keep_original: Whether to keep the original fields

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process predictions by restructuring JSON fields.

JSONKeyRenamerPostProcessor​

class JSONKeyRenamerPostProcessor(    column_patterns: list[str],    key_mappings: list[dict[str, str]],    recursive: bool = True,):

Renames keys within JSON data stored in columns.

Initialize the JSON key renamer.

Arguments

  • column_patterns: List of regex patterns matching columns to process
  • key_mappings: List of mappings, each with:
    • 'source_key': Original key name
    • 'target_key': New key name
  • recursive: Whether to rename keys recursively throughout nested objects

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process predictions by renaming JSON keys.

JSONWrapInListPostProcessor​

class JSONWrapInListPostProcessor(column_patterns: list[str]):

Wraps JSON data in an additional list layer.

Initialize the JSON wrap in list processor.

Arguments

  • column_patterns: List of regex patterns matching columns to process

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process predictions by wrapping JSON data in an additional list.

StringToJSONPostProcessor​

class StringToJSONPostProcessor(column_patterns: list[str]):

Converts string columns containing JSON data to actual JSON/dict objects.

Initialize the string to JSON converter.

Arguments

  • column_patterns: List of regex patterns matching column names to process

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process predictions by converting string columns to JSON objects.

TransformationApplierPostProcessor​

class TransformationApplierPostProcessor(    transformations: Union[list[Transformation], list[dict[str, Any]]],):

Applies transformations from the transformations module.

Initialize the transformation applier.

Arguments

  • transformations: List of transformations to apply. Can be either:
    • A list of Transformation objects (for programmatic use)
    • A list of dicts (from YAML config) that will be parsed into Transformation objects

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Process the predictions using the transformations.

Arguments

  • predictions: Model output to process

Returns Processed predictions

ZipFlattenCoordinatesPostProcessor​

class ZipFlattenCoordinatesPostProcessor(    source_column: str,    target_column: str,    slice_indices_path: str,    coordinates_path: str,    drop_source: bool = False,):

Zips slice indices with per-slice coordinate groups and flattens.

Reads a dict from a source column, extracts two parallel arrays (slice indices and coordinate groups), and produces a flat list of [slice_idx, coord_0, coord_1] triples written to a target column.

Each coordinate pair is prepended with its corresponding slice index and written as-is (no reordering).

Initialise the zip-flatten-coordinates postprocessor.

Arguments

  • source_column: Column containing the parsed dict.
  • target_column: Name of the output column for the flattened coordinate triples.
  • slice_indices_path: Dot-notation path to the slice indices array within the dict.
  • coordinates_path: Dot-notation path to the per-slice coordinate groups array within the dict.
  • drop_source: Whether to drop the source column after processing.

Ancestors​

Methods​

process​

def process(self, predictions: Any) ‑> Any:

Zip-flatten coordinates from JSON dicts into a new column.