Step 6: Analyze Model Performance

Use the code below to run TensorFlow Model Analysis on the model in your pipeline. Start by importing and opening the metadata store.



In [ ]:

    
from __future__ import print_function

import os
import tfx_utils

def _make_default_sqlite_uri(pipeline_name):
    return os.path.join(os.environ['HOME'], 'airflow/tfx/metadata', pipeline_name, 'metadata.db')

def get_metadata_store(pipeline_name):
    return tfx_utils.TFXReadonlyMetadataStore.from_sqlite_db(_make_default_sqlite_uri(pipeline_name))

pipeline_name = 'taxi' # or taxi_solution
pipeline_db_path = _make_default_sqlite_uri(pipeline_name)
print('Pipeline DB:\n{}'.format(pipeline_db_path))

store = get_metadata_store(pipeline_name)

Now print out the model artifacts:



In [ ]:

    
store.get_artifacts_of_type_df(tfx_utils.TFXArtifactTypes.MODEL)

Now analyze the model performance:



In [ ]:

    
store.display_tfma_analysis(<insert artifact ID here>, slicing_column='trip_start_hour')

Now plot the artifact lineage:



In [ ]:

    
# Try different IDs here. Click stop in the plot when changing IDs.
%matplotlib notebook
store.plot_artifact_lineage(<insert artifact ID here>)