[SC-8072] Support threshold lines in unit metric plots

[juanmleng]

Internal Notes for Reviewers

In addition to updates in the ValidMind library, supporting thresholds in log_metric() requires changes in both the backend and frontend repositories. The following pull requests have been created for these repos:

• https://github.com/validmind/backend/pull/1133
• https://github.com/validmind/frontend/pull/1176

External Release Notes

Description
Enhances the log_metric() function to accept multiple named thresholds as a dictionary input. This allows users to define and track multiple threshold levels (e.g., "high_risk": 0.6, "medium_risk": 0.7, "low_risk": 0.8) for each unit metric.

Changes

• Added threshold support in log_metric()
• Added notebook how_to\log_metrics_over_time.ipynb

Testing

• Tested with various threshold configurations
• Verified threshold line rendering in the UI
• Validated threshold updates reflect immediately in visualization

Example

log_metric(
    key="AUC Score",
    value=auc,
    recorded_at=datetime(2024, 1, 1),
    thresholds={
        "high_risk": 0.6,
        "medium_risk": 0.7,
        "low_risk": 0.8,
    }
)

[johnwalz97]

lgtm

[github-actions]

PR Summary

This pull request introduces enhancements to the ValidMind library by adding support for logging thresholds alongside metrics. The changes are primarily focused on the log_metric and alog_metric functions within the api_client.py file. These functions now accept an optional thresholds parameter, which allows users to specify threshold values for metrics. This feature is useful for visualizing metrics over time and identifying potential issues by setting reference lines in metric visualizations. The notebook log_metrics_over_time.ipynb has been updated to demonstrate how to use this new feature, including examples of logging multiple metrics with custom thresholds.

Test Suggestions

• Test logging a metric with valid threshold values to ensure they are correctly processed and stored.
• Test logging a metric without thresholds to verify that the function handles it gracefully.
• Test logging a metric with invalid threshold data types (e.g., a list instead of a dictionary) to ensure proper error handling.
• Verify that the notebook log_metrics_over_time.ipynb runs without errors and correctly logs metrics with thresholds.
• Check the visualization of metrics over time to ensure thresholds are displayed as expected.