In-memory + optional file-backed logger.
The default logger when a user does Experiment(log=True). Stores every event
in a list so the UI / notebook can replay it afterwards, and optionally tees
each event to a JSON-lines file.
Cheap, thread-safe enough for single-process use (the event list uses a list append which is atomic in CPython), and dependency-free.
Captures emitted events in an in-memory list and (optionally) a JSONL file.
Parameters
----------
experiment_id : str, optional
Stable id stamped on every event for correlating multi-experiment runs.
file : str | Path, optional
When provided, each event is also appended as a JSON line to this path.
The file is flushed on every write so a UI tailing it sees progress.
Usage
-----
>>> logger = MemoryLogger()
>>> exp = ClassificationExperiment(target="y", logger=logger).fit(data)
>>> exp.compare_models()
>>> logger.events # list of Event dataclasses
>>> logger.as_jsonl() # ""-joined JSONL string