api_reference.md

API Reference

The syntharc architecture relies heavily on a unified interface abstraction. All functional synthesizers inherit from BaseSynthesizer.

BaseSynthesizer (syntharc.core.base)

BaseSynthesizer(config: dict | None = None)

An abstract base class that enforces method chaining and the three-stage lifecycle. Base implementations automatically handle standard logging, config tracking, and boolean state checks (self.is_prepared, self.is_fitted).

Core Methods

• prepare(data=None, **kwargs) -> self: Handles hardware setups, model downloading, template extraction, and data caching. Must be called before generate() for models inheriting the _lifecycle = "prepare" class attribute.

• fit(data, **kwargs) -> self: Consumes pandas dataframes or python lists to adapt statistical curves, markov probabilities, or ML weights. Must be called before generate() for models inheriting the _lifecycle = "fit" class attribute.

• generate(num_samples: int, **kwargs) -> Any: Executes the generation loop using the prior prepared/fitted state. Parameter arguments vary by synthesizer type (e.g. seed, max_length, instructions).

• process(data, num_samples, **kwargs): Convenience method that wraps all three. Calls prepare(), fit(), and returns the exact output of generate().

Evaluation Utilities

Every syntharc module contains an evaluation.py file exposing functions to score the synthetic distributions:

• syntharc.tabular.evaluation.evaluate_tabular(real, synth)
• syntharc.image.evaluation.evaluate_images(real, synth)
• syntharc.text.evaluation.evaluate_text(real, synth)
• syntharc.timeseries.evaluation.evaluate_timeseries(real, synth)

These functions generally return a dictionary of numeric scores denoting distance, similarity bounds, or mathematical overlaps.