API Index¶

This page provides a complete reference to all classes, methods, and functions available in TwinWeaver. Click on any item to view its detailed documentation.

Quick Import¶

All main components can be imported directly from twinweaver:

from twinweaver import (
    # Core
    Config,
    DataManager,

    # Instruction Tuning
    ConverterInstruction,
    DataSplitter,
    DataSplitterForecasting,
    DataSplitterEvents,

    # Pretraining
    ConverterPretrain,

    # Utilities
    convert_meds_to_dtc,
    identify_constant_and_changing_columns,
    aggregate_events_to_weeks,
)

Common Module¶

Config¶

Configuration manager for TwinWeaver settings.

Member	Type	Description
`Config`	Class	Centralized configuration repository
`Config.set_delta_time_unit`	Method	Set the time unit for delta calculations
`Config.seed`	Property	Random seed for reproducibility

DataManager¶

Handles data loading and management.

Member	Type	Description
`DataManager`	Class	Manages data loading, processing, and splitting
`DataManager.load_indication_data`	Method	Load data tables for a specific indication
`DataManager.process_indication_data`	Method	Process loaded indication data
`DataManager.setup_unique_mapping_of_events`	Method	Create unique mapping for all events
`DataManager.setup_dataset_splits`	Method	Split data into train/val/test sets
`DataManager.get_all_patientids_in_split`	Method	Get all patient IDs in a specific split
`DataManager.get_patient_split`	Method	Get the split assignment for a patient
`DataManager.get_patient_data`	Method	Retrieve all data for a specific patient
`DataManager.infer_var_types`	Method	Infer variable types (numeric/categorical)

ConverterBase¶

Base class for all converters.

Member	Type	Description
`ConverterBase`	Class	Base class for data-to-text conversion
`round_and_strip`	Function	Format numbers with precision control
`ConverterBase.get_difference_in_event_dataframes`	Method	Compare two event DataFrames
`ConverterBase.forward_conversion_inference`	Method	Abstract method for inference conversion
`ConverterBase.generate_target_manual`	Method	Abstract method to generate target text
`ConverterBase.aggregate_multiple_responses`	Method	Abstract method to aggregate responses

Instruction Module¶

ConverterInstruction¶

Main converter for instruction-tuning data.

Member	Type	Description
`ConverterInstruction`	Class	Converter combining forecasting and events
`ConverterInstruction.set_custom_summarized_row_fn`	Method	Set custom function for row summarization
`ConverterInstruction.get_nr_tokens`	Method	Count tokens in a string
`ConverterInstruction.forward_conversion`	Method	Convert patient data to text (training)
`ConverterInstruction.forward_conversion_inference`	Method	Convert patient data to text (inference)
`ConverterInstruction.generate_target_manual`	Method	Generate target text from reverse conversion
`ConverterInstruction.aggregate_multiple_responses`	Method	Aggregate multiple LLM responses
`ConverterInstruction.reverse_conversion`	Method	Convert text back to structured data
`ConverterInstruction.get_difference_in_event_dataframes`	Method	Compare predicted vs actual events

ConverterForecasting¶

Converter for forecasting tasks.

Member	Type	Description
`ConverterForecasting`	Class	Converter for time-series forecasting
`ConverterForecasting.forward_conversion`	Method	Convert patient split to text (training)
`ConverterForecasting.forward_conversion_inference`	Method	Convert patient split to text (inference)
`ConverterForecasting.generate_target_manual`	Method	Generate forecasting target text
`ConverterForecasting.aggregate_multiple_responses`	Method	Aggregate multiple forecasting responses
`ConverterForecasting.reverse_conversion`	Method	Parse text back to structured forecasts

ConverterForecastingQA¶

Converter for forecasting with Q&A format.

Member	Type	Description
`ConverterForecastingQA`	Class	Q&A format forecasting converter
`ConverterForecastingQA.forward_conversion`	Method	Convert to Q&A format text

ConverterEvents¶

Converter for event prediction tasks.

Member	Type	Description
`ConverterEvents`	Class	Converter for clinical event prediction
`ConverterEvents.forward_conversion`	Method	Convert patient split to text (training)
`ConverterEvents.forward_conversion_inference`	Method	Convert patient split to text (inference)
`ConverterEvents.generate_target_manual`	Method	Generate event prediction target
`ConverterEvents.reverse_conversion`	Method	Parse text back to event predictions
`ConverterEvents.get_difference_in_event_dataframes`	Method	Compare predicted vs actual events
`ConverterEvents.aggregate_multiple_responses`	Method	Aggregate multiple event responses

DataSplitter¶

Main data splitter combining forecasting and events.

Member	Type	Description
`DataSplitter`	Class	Combined data splitter
`DataSplitter.get_splits_from_patient_with_target`	Method	Get splits with target data (training)
`DataSplitter.get_splits_from_patient_inference`	Method	Get splits for inference

DataSplitterForecasting¶

Splitter for forecasting data.

Member	Type	Description
`DataSplitterForecasting`	Class	Forecasting-specific data splitter
`DataSplitterForecasting.setup_statistics`	Method	Compute statistics for normalization
`DataSplitterForecasting.get_splits_from_patient`	Method	Generate splits for a patient
`DataSplitterForecastingOption`	Class	Configuration for forecasting splits
`DataSplitterForecastingGroup`	Class	Grouping for forecasting options
`DataSplitterForecastingGroup.append`	Method	Add option to group

DataSplitterEvents¶

Splitter for event data.

Member	Type	Description
`DataSplitterEvents`	Class	Event-specific data splitter
`DataSplitterEvents.setup_variables`	Method	Setup event variables
`DataSplitterEvents.get_splits_from_patient`	Method	Generate splits for a patient
`DataSplitterEventsOption`	Class	Configuration for event splits
`DataSplitterEventsGroup`	Class	Grouping for event options
`DataSplitterEventsGroup.append`	Method	Add option to group

BaseDataSplitter¶

Base class for all data splitters.

Member	Type	Description
`BaseDataSplitter`	Class	Abstract base for splitters
`BaseDataSplitter.select_random_splits`	Method	Randomly select split points
`BaseDataSplitter.drop_duplicates_except_na_for_date_col`	Method	Remove duplicates preserving NA dates

Pretrain Module¶

ConverterPretrain¶

Converter for pretraining data.

Member	Type	Description
`ConverterPretrain`	Class	Converter for pretraining data preparation
`ConverterPretrain.forward_conversion`	Method	Convert patient data to pretraining text
`ConverterPretrain.reverse_conversion`	Method	Parse pretraining text back to data

Utils Module¶

Utility functions for data preprocessing and integration.

Name	Type	Description
`convert_meds_to_dtc`	Function	Convert MEDS format data to TwinWeaver format
`identify_constant_and_changing_columns`	Function	Identify constant vs. time-varying columns
`aggregate_events_to_weeks`	Function	Aggregate event data to weekly intervals

Module Reference¶

For complete module documentation with full parameter details, see the API Reference section in the navigation.

Module	Description
`twinweaver.common`	Core configuration and base classes
`twinweaver.instruction`	Instruction-tuning converters and splitters
`twinweaver.pretrain`	Pretraining data converters
`twinweaver.utils`	Utility functions and integrations