Root Cause Service

Hydro serving models used in demos are stored in hs_demos.

You can test anchor and rise with python tests/test_anchor_on_multiple_adult_models.py and test/test_rise_on_mobilenet.py correspondingly.

Explanation State Diagram

Which models are eligible for creating an auto-od metric?

hydro-root-cause creates explanations for tabular or image tensors.

Right now only tabular explanations are available

If model signature is not supported, then there are no way to ask for explanation and explanation status will always be NOT_SUPPORTED

Model signature is considered supported when it consists of numerical scalar fields only. It should also have output field with name specified in config['explained_output_field_name']. If there are no such tensor, model signature will be considered NOT_SUPPORTED. You can change this tensor by update config during runtime, and model status will be refreshed.

Supported fields are:

• of scalar shape
• of types:
  • DT_FLOAT
  • DT_DOUBLE
  • DT_INT8
  • DT_INT16
  • DT_INT32
  • DT_INT64

In future more model fields will be supported.

Environment variables to configure service while deploying

Addresses to other services:

• HTTP_UI_ADDRESS - http address of hydro-serving cluster, used to create hydrosdk.Cluster(HS_CLUSTER_ADDRESS)
• GRPC_UI_ADDRESS - grpc address

MongoDB parameters:

• MONGO_URL
• MONGO_PORT
• MONGO_AUTH_DB
• MONGO_USER
• MONGO_PASS
• ROOT_CAUSE_DB_NAME - Name of database in mongo which will be used for this service

AWS/Minio parameters:

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• S3_ENDOPOINT, if no s3 endpoint is present, data will be downloaded with boto3

Flask server parameters:

• APPLICATION_ROOT - prefix of all routes specified in hydro_auto_od_openapi.yaml, not used right now
• DEBUG