This is the open source cloudexplain package (or what it is going to be). This just uploads your current data to the azure account you are currently logged in via az.
To install this for azure run:
pip install cloudexplain[azure]
In order to run it on azure you must have write rights on the cloudexplainmodels storage account on the resource group cloudexplain on your subscription and must be logged in via
az login
In training mode you need to specify the y argument.
import shap import sklearn import cloudexplain X, y = shap.datasets.adult(n_points=100) X["adult_id"] = list(range(len(X))) dtc = sklearn.tree.DecisionTreeClassifier() dtc.fit(X, y) with cloudexplain.azure.explain(model=dtc, X=X, y=y, model_version="0.0.1", model_description="This is a new model.", resource_group_name="mycomp-cloudexplain-tf", explanation_name="dummy_name", explanation_env="dev", data_source="shap_adult_dataset", observation_id_column="adult_id") as run: print("This is the run uuid", run.run_uuid) pred = dtc.predict(X)
To run in inference mode, just don't specify y.
with cloudexplain.azure.explain(model=dtc, X=X, model_version="0.0.1", model_description="This is a new model.", resource_group_name="mycomp-cloudexplain-tf", explanation_name="dummy_name", explanation_env="dev", data_source="shap_adult_dataset", observation_id_column="adult_id") as run: print("This is the run uuid", run.run_uuid) pred = dtc.predict(X)
When logged in with multiple accounts it can be the case that the permissions are not handled correctly. Signing out of all users except for the one who has the correct rights helps.
az logoutIf AuthorizationPermissionMismatch is raised for a newly spun up cloudexplain resource, try logging out and in again. This refreshes azure privileges and should resolve the issue.