Library Reference¶

Core Functions¶

Function	Purpose	Returns
`adapt_route(raw, adapter)`	Adapt one raw route-like payload	`Route \\| None`
`adapt_provider_output(raw, adapter)`	Adapt one raw provider output	`list[Route]`
`adapt_target_keyed_provider_output(raw, benchmark, adapter)`	Adapt raw output keyed by target id or smiles	`list[Route]`
`adapt_routes(raw, target, adapter)`	Deprecated in v0.6; use provider-output workflows	`list[Route]`
`adapt_single_route(raw, target, adapter)`	Deprecated in v0.6; use `adapt_route(raw, adapter)`	`Route \\| None`
`collect_benchmark_predictions(routes, benchmark)`	Collect canonical routes onto benchmark targets	`CollectedBenchmarkRoutes`
`deduplicate_routes(routes)`	Remove duplicate routes	`list[Route]`
`score_predictions(model_name, benchmark, predictions, stock)`	Evaluate routes	`ScoredResults`
`compute_model_statistics(results, n_boot)`	Bootstrap statistics	`ModelStatistics`
`load_benchmark(path)`	Load benchmark definition	`Benchmark`
`load_stock_file(path)`	Load stock molecules	`set[str]`

Visualization Functions¶

Function	Purpose	Returns
`plot_diagnostics(stats)`	Single model performance	`plotly.Figure`
`plot_comparison(models_stats, metric_type, k)`	Multi-model comparison	`plotly.Figure`

Adapters¶

from retrocast.adapters import ADAPTER_MAP

for name in ADAPTER_MAP:
    print(name)

Supported adapters:

aizynth, askcos, dms, dreamretro, molbuilder, multistepttl, paroutes, retrochimera, retrostar, synllama, synplanner, syntheseus, ursa-llm