Python SDK Reference

The inferpathio package provides a high-level interface for experiment tracking, model registration, and drift monitoring. Requires Python 3.8+.

Installation

pip install inferpathio

ifp.track()

Context manager that creates a new experiment run and ensures it's closed on exit.

ifp.track(model_name: str, run_name: str = None, tags: list = None)
  • model_name — Logical name for the model being trained
  • run_name — Optional human-readable label for this run
  • tags — Optional list of string tags for filtering

run.log_metric()

Log a scalar metric at a given step.

run.log_metric(key: str, value: float, step: int = None)

run.log_param()

Log a hyperparameter value. Values can be strings, numbers, or booleans.

run.log_param(key: str, value)

run.register()

Register the trained model artifact in the registry.

run.register(
  stage: str = "staging",
  artifact_path: str = None,
  tags: list = None
)

ifp.monitor()

Configure a drift monitor for a registered model.

m = ifp.monitor("model-name")
m.set_baseline(X_train)
m.drift_threshold = 0.10
m.on_drift(action="alert", channel="slack:#ml-ops")
m.watch(X_live)

Framework adapters

PyTorch

from inferpathio.adapters import pytorch as ifp_torch
callback = ifp_torch.IFPCallback(run)
trainer.add_callback(callback)

scikit-learn

from inferpathio.adapters import sklearn as ifp_sk
ifp_sk.patch(run)  # monkey-patches fit() to auto-log metrics