Module

pycaret.logging.events

Structured event types emitted by PyCaret 4.0 operations.

Events are the single source of truth for "what happened during this run":

Every Experiment method emits at least one event.
Events are immutable dataclasses that can be serialized to JSON.
A React UI subscribes to the stream (via BaseLogger.subscribe(...)) and renders progress; an LLM agent reads the same stream as trace data.

EventKind is a string-enum so events can round-trip through JSON without custom encoders.

Classes 2

extends StrEnum

Canonical event kinds. String-valued for JSON round-tripping.

Attributes

An immutable record of something that happened during an experiment.

kind : EventKind

Canonical kind string.

timestamp : float

Unix seconds since the epoch, captured via `time.time()`.

duration_ms : float, optional

Wall-clock duration for "finished" events. `None` for point events.

message : str

Short human-readable summary.

payload : dict

Structured fields specific to this event kind (e.g. `{"model_id": "lr",
"score": 0.83}`). JSON-serializable by convention.

experiment_id : str, optional

Stable id of the emitting experiment (defaults to the experiment's
`session_id` stringified).

Attributes

Methods

to_dict(self) -> dict[str, Any]

Return a JSON-serializable dict representation.