MetaflowOfflineTrain¶

The MetaflowOfflineTrain class is a subclass of BaseTrain and provides methods for running a Metaflow pipeline and retrieving artifacts associated with the pipeline run.

Attributes¶

METAFLOW_CLASS - The name of the class that inherits from metaflow's FlowSpec

Configuration¶

Required Configuration¶

The MetaflowOfflineTrain requires the following components:

data_splitter
metadata_tracker
resource_version_control
model_explainer
model_checker
model_visualizer
model_bias_checker

Methods¶

run¶

The run() method executes the Metaflow pipeline with the provided data. It takes a pandas dataframe data as input, which represents the data to be used in the training process.

def run(self, data, *args, **kwargs)

Arguments:

data (dict): The dictionary of train/test/validation data, returned by the data_splitter component.

Returns

None

get_artifacts¶

he get_artifacts() method retrieves the artifacts associated with the Metaflow pipeline run. It takes a list of artifact keys artifact_keys as input, representing the artifacts to retrieve from the run. The method returns a dictionary artifacts containing the retrieved artifacts.

def get_artifacts(self, artifact_keys):

Arguments:

artifact_keys (list): A list of artifact keys to retrieve.

Returns

artifacts (dict): A dictionary containing the requested artifacts.

MetaflowOfflineTrainSpec Methods¶

MetaflowOfflineTrainSpec contains the following methods. These are mirrored from the OfflineTrain class, and you should see that documentation for more information (Note: instead of these method explicitly using arguments, they instead access saved artifacts during the Metaflow run).

start
split_data
train_model
check_model
analyze_model
compare_models
check_model_bias
retrain_model_on_all_data
end