src.history_managing module

Save pipeline history artifacts and results.

This module provides functions for saving and formatting pipeline artifacts such as LLM generation history, generated images, and experiment parameters.

src.history_managing.format_as_json_string(dictionaries_list: Sequence[Mapping[str, Any]]) → Sequence[str]

Convert a list of dictionaries to JSON-formatted strings.

Parameters:

dictionaries_list – Sequence of dictionaries to format.

Returns:

Sequence of JSON-formatted strings, one per dictionary.

src.history_managing.format_best_concepts_history(best_concept_history: Sequence[Mapping[str, Any]]) → Sequence[str]

Format best concepts history as JSON-formatted strings.

Parameters:

best_concept_history – Sequence of best concept dictionaries.

Returns:

Sequence of JSON-formatted strings for best concepts.

src.history_managing.format_concept_history(concept_history: Mapping[str, float]) → Sequence[str]

Format concept history as comma-separated concept-score strings.

Parameters:

concept_history – Dictionary mapping concept names to scores.

Returns:

Sequence of formatted strings with format ‘concept,score’.

src.history_managing.save_images_from_iteration(images: Sequence[PIL.Image.Image], save_directory: str, iter_number: int, concept: str) → None

Save images generated in a single pipeline iteration.

Creates a subdirectory for the iteration and saves each image as a JPEG file.

Parameters:

• images – Sequence of PIL Image objects to save.

• save_directory – Base directory to save images to.

• iter_number – Iteration number for organization.

• concept – Concept name, used in directory naming.

src.history_managing.save_llm_history(history: Iterable[str], save_directory: str, filename: str) → None

Save language model generation history to a file.

Parameters:

• history – Iterable of generation history strings.

• save_directory – Directory to save the history file to.

• filename – Name of the file to save.

src.history_managing.save_pipeline_parameters(save_directory: str, run_id: str, load_config: LoadConfig, image_generation_config: ImageGenerationConfig, concept_history_config: ConceptHistoryConfig, history_managing_config: HistoryManagingConfig, neuron_id: int, metric: Metric, model_layer_activations_path: str)

Save all pipeline configuration parameters to a file.

Creates a formatted text file with all experiment parameters including model configurations, generation settings, and neuron information.

Parameters:

• save_directory – Directory to save parameters to.

• run_id – Unique identifier for this pipeline run.

• load_config – Model loading configuration.

• image_generation_config – Image generation settings.

• concept_history_config – Concept history initialization settings.

• history_managing_config – Result saving configuration.

• neuron_id – ID of the neuron being explained.

• metric – Activation metric being used.

• model_layer_activations_path – Path to control concept activations.