Library Reference¶

This page lists the main schema-2 entry points. See the neighboring guides for context and examples.

Adaptation¶

| Function | Purpose | Returns | | --- | --- | --- | --- | | adapt_route(raw_route, adapter) | Adapt one raw route record | Route | None | | adapt_routes(raw_payload, adapter) | Adapt a raw artifact and keep only successful routes | list[Route] | | adapt_candidates(raw_payload, adapter) | Adapt a raw artifact while preserving failed prediction slots | list[Candidate] | | collect_candidates(candidates, task) | Map candidates onto task targets | dict[str, list[Candidate]] | | collect_routes(routes, task) | Map route-only outputs onto task targets | dict[str, list[Route]] | | ingest_candidates(raw_payload, adapter, task) | adapt_candidates + collect_candidates | dict[str, list[Candidate]] | | ingest_routes(raw_payload, adapter, task) | adapt_routes + collect_routes | dict[str, list[Route]] |

Scoring and Analysis¶

Function	Purpose	Returns
`score(predictions, task, constraint_checkers=...)`	Score collected candidates	`Evaluation`
`score_candidate(candidate, target=..., constraints=..., ...)`	Score one candidate	`ScoredCandidate`
`score_target(candidates, target=..., constraints=..., ...)`	Score one target's candidates	`TargetResult`
`analyze(evaluation, ks=(1, 5, 10, 50), n_boot=10000)`	Summarize an evaluation	`AnalysisReport`
`score_predictions(predictions, task, stocks=...)`	Convenience wrapper that builds default task constraint checkers	`Evaluation`
`analyze_evaluation(evaluation, n_boot=10000)`	Convenience wrapper around `analyze(...)`	`AnalysisReport`

IO¶

Function	Purpose	Shape
`load_benchmark(path)` / `save_benchmark(...)`	Benchmark artifacts	`Benchmark`
`load_candidates(path)` / `save_candidates(...)`	Flat candidate artifacts	`list[Candidate]`
`load_collected_candidates(path)` / `save_collected_candidates(...)`	Collected candidate artifacts	`dict[str, list[Candidate]]`
`load_evaluation(path)` / `save_evaluation(...)`	Scored artifacts	`Evaluation`
`load_analysis_report(path)` / `save_analysis_report(...)`	Analysis artifacts	`AnalysisReport`
`load_stock_file(path)`	Stock files	`set[InChIKeyStr]` or `set[SmilesStr]`

Core Models¶

Model	Purpose
`Route`, `Molecule`, `Reaction`	Canonical route tree
`RoutePath`, `ReactionId`, `MoleculeId`	Route-local addressing
`Candidate`, `FailureRecord`	Adaptation accounting
`Target`, `TaskConstraint`, `Task`, `Benchmark`	Problem definition
`CheckResult`, `TierResult`, `RouteValidity`, `ConstraintResult`	Scoring details
`ScoredCandidate`, `TargetResult`, `Evaluation`	Scored prediction artifact
`MetricSummary`, `AnalysisReport`	Analysis artifact

Adapters¶

from retrocast.adapters import ADAPTER_TYPES, get_adapter

print(sorted(ADAPTER_TYPES))
adapter = get_adapter("paroutes")

Supported adapters include aizynthfinder, askcos, directmultistep, dreamretroer, molbuilder, multistepttl, paroutes, retrochimera, retrostar, synllama, synplanner, syntheseus, and ursa.