Configuration¶

This page describes how to configure the tracking module that forwards the pipeline's output to a tracking backend.

Options¶

Name	Allowed	Default	Description
`_target_`	`"MLFlowTracker"`	`"MLFlowTracker"`	Hydra target for the tracking backend implementation.
`output_forwarding_url`	`string, null`	`null`	Tracking URI used by the backend. For `MLFlowTracker`, this is passed to MLflow as the tracking URI. It can point to a local path or an existing HTTP tracking server. If not set, it uses `sqlite:///mlflow/mlflow.db`, with the database at `mlflow/mlflow.db` from the repository root and artifacts under `mlflow/artifacts`.
`backend_store_uri`	`string, null`	`null`	MLflow backend store URI used when RAITAP starts a local MLflow server, or when `output_forwarding_url` is not set. For `MLFlowTracker`, an unset value falls back to `sqlite:///mlflow/mlflow.db`. SQLite URIs are supported for local tracking storage.
`default_artifact_root`	`string, null`	`null`	MLflow artifact root used when RAITAP manages the local MLflow backend store or starts a local MLflow server. For `MLFlowTracker`, an unset value falls back to `./mlflow/artifacts` when RAITAP owns the local backend. If `output_forwarding_url` points to an explicit local file store and this option is not set, MLflow's default artifact location is used.
`log_model`	`boolean`	`false`	Whether to log the assessed model to the tracking backend. Note that this might take significant time and resources for large models.
`open_when_done`	`boolean`	`true`	Whether to open the tracking UI automatically after the run completes.

YAML

tracking:
  _target_: "MLFlowTracker"
  output_forwarding_url: "http://127.0.0.1:5001"
  log_model: true

Python

from raitap.tracking import mlflow

tracking = mlflow(
    output_forwarding_url="http://127.0.0.1:5001",
    log_model=True,
)

CLI

uv run raitap +tracking=mlflow tracking.log_model=true

pip

raitap +tracking=mlflow tracking.log_model=true