How to access the model metrics and features from a deployed model?

Options
Mohammed
Mohammed Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Registered Posts: 39 ✭✭✭

Hi,
I have deployed a model in Dataiku flow. I'm trying to access the model from a notebook and get some model details, as shown below.

import dataiku# Connect to the Dataiku instanceclient = dataiku.api_client()project = client.get_project(PROJECT_KEY)# Access the saved modelmodel_id = 'MODEL_ID'saved_model = project.get_saved_model(model_id)active_version_id=saved_model.get_active_version()['id']saved_model.get_metric_values(active_version_id)

But instead of returning a list of metrics, it returns the following.

<dataikuapi.dss.metrics.ComputedMetrics at 0x7ff5ddf5eee0>


How do I access the model metrics? Similarly, how do I access the features used in the model?


Operating system used: Windows


Operating system used: Windows

Answers

  • AdrienL
    AdrienL Dataiker, Alpha Tester Posts: 196 Dataiker
    Options

    If you have a ComputedMetrics object, you can for instance call its get_all_ids method to list the metrics ids, then its get_global_value method to get the value for a given metric.

    However, for a Saved Model's active version, it may be simpler, for what it seems you're trying to achieve, to get the metrics via get_version_details that returns a DSSTrainedPredictionModelDetails object, on which you can call get_performance_metrics for instance.

  • Mohammed
    Mohammed Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Registered Posts: 39 ✭✭✭
    Options

    What about the features? How do I get the features used in a model that is present in the lab (not saved) ?

  • AdrienL
    AdrienL Dataiker, Alpha Tester Posts: 196 Dataiker
    Options

    For the feature, the train model details can also give the preprocessing settings containing those.

    features_handling = details.get_preprocessing_settings()['per_feature']features_used = {name: (features_handling[name]['role']=='INPUT') for name in features_handling}

    features_used will have feature names as keys and a boolean as value, with True when it was used as an input to the model's training and False if it wasn't.

Setup Info
    Tags
      Help me…