microsoft/qlib · error · NotImplementedError

Please implement the `start_exp` method.

Error message

Please implement the `start_exp` method.

What it means

ExpManager is qlib's abstract experiment-manager base class; the public start_exp handles bookkeeping (active URI) then delegates to _start_exp, whose base implementation raises NotImplementedError. Only concrete managers (MLflowExpManager) implement it, so this error means the abstract manager, or a subclass missing _start_exp, was used to start an experiment.

Source

Thrown at qlib/workflow/expm.py:96

        Returns
        -------
        An active experiment.
        """
        self._active_exp_uri = uri
        # The subclass may set the underlying uri back.
        # So setting `_active_exp_uri` come before `_start_exp`
        return self._start_exp(
            experiment_id=experiment_id,
            experiment_name=experiment_name,
            recorder_id=recorder_id,
            recorder_name=recorder_name,
            resume=resume,
            **kwargs,
        )

    def _start_exp(self, *args, **kwargs) -> Experiment:
        """Please refer to the doc of `start_exp`"""
        raise NotImplementedError(f"Please implement the `start_exp` method.")

    def end_exp(self, recorder_status: Text = Recorder.STATUS_S, **kwargs):
        """
        End an active experiment.

        Maintaining `_active_exp_uri` is included in end_exp, remaining implementation should be included in _end_exp in subclass

        Parameters
        ----------
        experiment_name : str
            name of the active experiment.
        recorder_status : str
            the status of the active recorder of the experiment.
        """
        self._active_exp_uri = None
        # The subclass may set the underlying uri back.
        # So setting `_active_exp_uri` come before `_end_exp`
        self._end_exp(recorder_status=recorder_status, **kwargs)

View on GitHub (pinned to 79633dd950)

Solutions

1. Restore the default: C.exp_manager = {'class': 'MLflowExpManager', 'kwargs': {'uri': ...}} in qlib.init.
2. In your custom manager subclass, implement _start_exp(experiment_id, experiment_name, recorder_id, recorder_name, resume, **kwargs) returning an Experiment.
3. Verify the configured class actually subclasses ExpManager AND overrides the private hooks, not just the public methods.

Example fix

# before
qlib.init(exp_manager={'class': 'ExpManager', 'kwargs': {'uri': 'file:./mlruns'}})
R.start_exp()  # NotImplementedError

# after
qlib.init()  # defaults to MLflowExpManager
R.start_exp(record_name='test')

from qlib.workflow.expm import ExpManager
assert type(R.exp_manager) is not ExpManager and R.exp_manager.__class__._start_exp is not ExpManager._start_exp

def manager_can_start(mgr) -> bool:
    from qlib.workflow.expm import ExpManager
    return mgr.__class__._start_exp is not ExpManager._start_exp

When it happens

Trigger: R.start_exp(...) after C['exp_manager']['class'] was set to the base 'ExpManager' or to a custom class that does not override _start_exp; directly instantiating ExpManager and calling start_exp.

Common situations: Customizing qlib config to point at a user-written tracking manager that is still a skeleton; typos in the configured class name that happen to resolve to the base class; code written against an internal fork missing the method.

Related errors

• Please implement the `end_exp` method.
• Please implement the `create_exp` method.
• Please implement the `search_records` method.
• Please implement the `_get_exp` method
• Please implement the `delete_exp` method.

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/3dcc1dfd5e3baf65. Report an issue: GitHub.