Load_Reference_Profile_from_WhyLabs.md

Loading a Reference Profile from WhyLabs

This guide explains how to use the load_reference_data_from_whylabs function in your project to load a reference profile from WhyLabs. The function fetches the profile using the WhyLabs API and provides it as a DatasetProfileView object for use in data validation and monitoring.

---

Prerequisites

1. Environment Variables

To use the function, you need to set the following environment variables:

• WHYLABS_DEFAULT_ORG_ID: Your WhyLabs organization ID.
• WHYLABS_DEFAULT_DATASET_ID: Your WhyLabs dataset (model) ID.
• WHYLABS_API_KEY: Your WhyLabs API key for authentication.
• WHYLABS_REF_ID: The unique reference profile ID.

WHYLABS_API_KEY

1. Go to the WhyLabs Hub.
2. Navigate to Menu → Settings → Manage API tokens.
3. Create or copy an API token for use in your project.

WHYLABS_REF_ID

1. Open your WhyLabs Dashboard.
2. Locate the reference profile you want to use.
3. Refer to the screenshot below:

For more information about working with environment variables in Valohai, refer to the Valohai Environment Variables Documentation.

2. Install Required Dependencies

Ensure the following libraries are installed:

pip install whylabs-client whylogs

3. Specify the Inference Step Parameter

When running the inference step, specify the use_whylabs_reference_profile parameter:

• use_whylabs_reference_profile: Indicates whether to load the reference profile from WhyLabs. If set to False, the reference profile will be created dynamically using the input data provided in the ref_data directory.

---

Function Explanation

The load_reference_data_from_whylabs function performs the following steps:

1. Environment Variable Loading:

   • The function retrieves ORG_ID, MODEL_ID, API_KEY, and REF_ID from the environment. If any are missing, it raises an error.

2. API Client Configuration:

   • Configures the WhyLabs API client using whylabs_client.Configuration. The API key is used for authentication.

3. Fetching Metadata:

   • Calls get_reference_profile from DatasetProfileApi to retrieve metadata for the reference profile.

4. Downloading the Profile:

   • Uses the download_url from the API response to fetch the reference profile as binary data.

5. Deserializing the Profile:

   • Converts the binary data into a DatasetProfileView object using whylogs.DatasetProfileView.deserialize.

6. Returning the Profile:

   • The function returns the reference_profile, ready for use in your pipeline.

---

References

1. WhyLabs Documentation:

   • For more details on managing profiles in WhyLabs, see the WhyLabs Documentation.

2. WhyLabs Python Client:

   • Learn more about the get_reference_profile API in the WhyLabs Python Client GitHub Documentation.

3. Valohai Environment Variables:

   • Learn how to manage environment variables in Valohai from the official documentation.

---

Example Usage

from your_project_module import load_reference_data_from_whylabs

# Ensure the required environment variables are set
reference_profile = load_reference_data_from_whylabs()
print(f"Successfully loaded reference profile with timestamp: {reference_profile.dataset_timestamp}")

When running inference, use the use_whylabs_reference_profile parameter to specify whether the reference profile is loaded from WhyLabs. You can adjust this parameter in the valohai.yaml file or through the Valohai UI.