mlflow.shap

mlflow.shap.get_default_conda_env()[source]
Returns

The default Conda environment for MLflow Models produced by calls to save_explainer() and log_explainer().

mlflow.shap.get_underlying_model_flavor(model)[source]

Find the underlying models flavor.

Parameters

model – underlying model of the explainer.

mlflow.shap.load_explainer(model_uri)[source]

Note

Experimental: This method may change or be removed in a future release without warning.

Load a SHAP explainer from a local file or a run.

Parameters

model_uri

The location, in URI format, of the MLflow model, for example:

  • /Users/me/path/to/local/model

  • relative/path/to/local/model

  • s3://my_bucket/path/to/model

  • runs:/<mlflow_run_id>/run-relative/path/to/model

  • models:/<model_name>/<model_version>

  • models:/<model_name>/<stage>

For more information about supported URI schemes, see Referencing Artifacts.

Returns

A SHAP explainer.

mlflow.shap.log_explainer(explainer, artifact_path, serialize_model_using_mlflow=True, conda_env=None, registered_model_name=None, signature: Optional[mlflow.models.signature.ModelSignature] = None, input_example: Optional[Union[pandas.core.frame.DataFrame, numpy.ndarray, dict, list]] = None, await_registration_for=300)[source]

Note

Experimental: This method may change or be removed in a future release without warning.

Log an SHAP explainer as an MLflow artifact for the current run.

Parameters
  • explainer – SHAP explainer to be saved.

  • artifact_path – Run-relative artifact path.

  • serialize_model_using_mlflow – When set to True, MLflow will extract the underlying model and serialize it as an MLmodel, otherwise it uses SHAP’s internal serialization. Defaults to True. Currently MLflow serialization is only supported for models of ‘sklearn’ or ‘pytorch’ flavors.

  • conda_env

    Either a dictionary representation of a Conda environment or the path to a Conda environment yaml file. If provided, this decsribes the environment this model should be run in. At minimum, it should specify the dependencies contained in get_default_conda_env(). If None, the default get_default_conda_env() environment is added to the model. The following is an example dictionary representation of a Conda environment:

    {
        'name': 'mlflow-env',
        'channels': ['defaults'],
        'dependencies': [
            'python=3.6.0',
            'shap=0.37.0'
        ]
    }
    

  • registered_model_name – (Experimental) If given, create a model version under registered_model_name, also creating a registered model if one with the given name does not exist.

  • signature

    (Experimental) ModelSignature describes model input and output Schema. The model signature can be inferred from datasets with valid model input (e.g. the training dataset with target column omitted) and valid model output (e.g. model predictions generated on the training dataset), for example:

    from mlflow.models.signature import infer_signature
    train = df.drop_column("target_label")
    predictions = ... # compute model predictions
    signature = infer_signature(train, predictions)
    

  • input_example – (Experimental) Input example provides one or several instances of valid model input. The example can be used as a hint of what data to feed the model. The given example will be converted to a Pandas DataFrame and then serialized to json using the Pandas split-oriented format. Bytes are base64-encoded.

  • await_registration_for – Number of seconds to wait for the model version to finish being created and is in READY status. By default, the function waits for five minutes. Specify 0 or None to skip waiting.

mlflow.shap.log_explanation(predict_function, features, artifact_path=None)[source]

Note

Experimental: This method may change or be removed in a future release without warning.

Given a predict_function capable of computing ML model output on the provided features, computes and logs explanations of an ML model’s output. Explanations are logged as a directory of artifacts containing the following items generated by SHAP (SHapley Additive exPlanations).

  • Base values

  • SHAP values (computed using shap.KernelExplainer)

  • Summary bar plot (shows the average impact of each feature on model output)

Parameters
  • predict_function

    A function to compute the output of a model (e.g. predict_proba method of scikit-learn classifiers). Must have the following signature:

    def predict_function(X) -> pred:
        ...
    
    • X: An array-like object whose shape should be (# samples, # features).

    • pred: An array-like object whose shape should be (# samples) for a regressor or (# classes, # samples) for a classifier. For a classifier, the values in pred should correspond to the predicted probability of each class.

    Acceptable array-like object types:

    • numpy.array

    • pandas.DataFrame

    • shap.common.DenseData

    • scipy.sparse matrix

  • features

    A matrix of features to compute SHAP values with. The provided features should have shape (# samples, # features), and can be either of the array-like object types listed above.

    Note

    Background data for shap.KernelExplainer is generated by subsampling features with shap.kmeans. The background data size is limited to 100 rows for performance reasons.

  • artifact_path – The run-relative artifact path to which the explanation is saved. If unspecified, defaults to “model_explanations_shap”.

Returns

Artifact URI of the logged explanations.

Example
import os

import numpy as np
import pandas as pd
from sklearn.datasets import load_boston
from sklearn.linear_model import LinearRegression

import mlflow

# prepare training data
dataset = load_boston()
X = pd.DataFrame(dataset.data[:50, :8], columns=dataset.feature_names[:8])
y = dataset.target[:50]

# train a model
model = LinearRegression()
model.fit(X, y)

# log an explanation
with mlflow.start_run() as run:
    mlflow.shap.log_explanation(model.predict, X)

# list artifacts
client = mlflow.tracking.MlflowClient()
artifact_path = "model_explanations_shap"
artifacts = [x.path for x in client.list_artifacts(run.info.run_id, artifact_path)]
print("# artifacts:")
print(artifacts)

# load back the logged explanation
dst_path = client.download_artifacts(run.info.run_id, artifact_path)
base_values = np.load(os.path.join(dst_path, "base_values.npy"))
shap_values = np.load(os.path.join(dst_path, "shap_values.npy"))

print("\n# base_values:")
print(base_values)
print("\n# shap_values:")
print(shap_values[:3])
Output
# artifacts:
['model_explanations_shap/base_values.npy',
 'model_explanations_shap/shap_values.npy',
 'model_explanations_shap/summary_bar_plot.png']

# base_values:
20.502000000000002

# shap_values:
[[ 2.09975523  0.4746513   7.63759026  0.        ]
 [ 2.00883109 -0.18816665 -0.14419184  0.        ]
 [ 2.00891772 -0.18816665 -0.14419184  0.        ]]
../_images/shap-ui-screenshot.png

Logged artifacts

mlflow.shap.save_explainer(explainer, path, serialize_model_using_mlflow=True, conda_env=None, mlflow_model=None, signature: Optional[mlflow.models.signature.ModelSignature] = None, input_example: Optional[Union[pandas.core.frame.DataFrame, numpy.ndarray, dict, list]] = None)[source]

Note

Experimental: This method may change or be removed in a future release without warning.

Save a SHAP explainer to a path on the local file system. Produces an MLflow Model containing the following flavors:

Parameters
  • explainer – SHAP explainer to be saved.

  • path – Local path where the explainer is to be saved.

  • serialize_model_using_mlflow – When set to True, MLflow will extract the underlying model and serialize it as an MLmodel, otherwise it uses SHAP’s internal serialization. Defaults to True. Currently MLflow serialization is only supported for models of ‘sklearn’ or ‘pytorch’ flavors.

  • conda_env

    Either a dictionary representation of a Conda environment or the path to a Conda environment yaml file. If provided, this decsribes the environment this model should be run in. At minimum, it should specify the dependencies contained in get_default_conda_env(). If None, the default get_default_conda_env() environment is added to the model. The following is an example dictionary representation of a Conda environment:

    {
        'name': 'mlflow-env',
        'channels': ['defaults'],
        'dependencies': [
            'python=3.6.0',
            'shap=0.37.0'
        ]
    }
    

  • mlflow_modelmlflow.models.Model this flavor is being added to.

  • signature

    (Experimental) ModelSignature describes model input and output Schema. The model signature can be inferred from datasets with valid model input (e.g. the training dataset with target column omitted) and valid model output (e.g. model predictions generated on the training dataset), for example:

    from mlflow.models.signature import infer_signature
    train = df.drop_column("target_label")
    predictions = ... # compute model predictions
    signature = infer_signature(train, predictions)
    

  • input_example – (Experimental) Input example provides one or several instances of valid model input. The example can be used as a hint of what data to feed the model. The given example will be converted to a Pandas DataFrame and then serialized to json using the Pandas split-oriented format. Bytes are base64-encoded.