Configuration and I/O¶

schema ¶

Experiment configuration files (YAML) and model references.

Configuration is data only: files are parsed with yaml.safe_load and validated against a strict schema (unknown keys are errors). The one way a config can lead to code running is the model: reference, which names a Python model explicitly - the same trust model as pointing pytest at a test file. Model files referenced by path must live inside the config file's directory unless the caller opts out, which prevents ../../ tricks.

loaded_model_files ¶

loaded_model_files() -> dict[str, str]

Module name -> path of model files loaded by path (re-loaded in worker processes).

preload_model_files ¶

preload_model_files(files: Mapping[str, str]) -> None

Worker-process initializer: import user model files under the same module names.

resolve_model ¶

resolve_model(ref: str, *, base_dir: str | Path | None = None, allow_outside: bool = True) -> Model

Resolve builtin:<name>, path/to/file.py[:attr] or package.module:attr.

With allow_outside=False file references must resolve inside base_dir.

build_experiment ¶

build_experiment(config: ExperimentConfig, *, base_dir: str | Path | None = None, allow_outside: bool = False) -> Any

Turn a validated config into a ready-to-run :class:~simulsi.Experiment.

io ¶

Safe, dependency-light export and import of results.

JSON is always available. NaN and infinities are written as null so files stay valid JSON for other tools (browsers reject bare NaN).
CSV uses the standard library.
Parquet needs the optional pyarrow + pandas extras.

Loading never uses pickle or eval: only JSON is read back.

to_jsonable ¶

to_jsonable(obj: Any) -> Any

Convert numpy scalars/arrays, dataclasses, tuples and non-finite floats for JSON.

csv_text ¶

csv_text(rows: Sequence[Mapping[str, Any]]) -> str

Dict rows as CSV text; the header is the union of keys in first-seen order.

write_csv ¶

write_csv(rows: Sequence[Mapping[str, Any]], path: str | Path) -> Path

Write dict rows; the header is the union of keys in first-seen order.

safe_child ¶

safe_child(base: str | Path, name: str) -> Path

Resolve base/name and refuse paths that escape base (path traversal).