Configure your First project

This guide uses a Regression model as an example to demonstrate the setup process.

Step 1: Cold Start

Figure 1. Data Mapping and Pipeline Configuration

Select your Problem Type - Regression.
Choose your Reference Pipeline and its associated dataframe. This is usually your test set from the training phase and must include:
- Features
- Model predictions
- Ground truth labels
Choose your Monitored Pipeline and its associated dataframe. This is your production data.

Schema Validity: The monitored dataframe must have a schema identical to the reference dataframe - feature columns, model predictions, ground truth column even if empty. Discrepancies will result in failure.

Step 2: Chunk Preferences

Specify how you want to aggregate your data.

Figure 2. Chunk Configuration Interface

Select a Timestamp column.
Select a Grouping Period (Hour, Day, Week, Month, Quarter, or Year).

Choosing the right grouping: Small chunks provide faster alerts but may result in noise or false positives. Larger chunks provide more stable statistical results but may delay detection of a sudden model failure.

Step 3: Performance Monitoring (Optional)

Figure 3. Performance Monitoring Configuration

Identify the Features used during training.
Map the Target Column (ground truth) and the Prediction Column.
Select one or more Performance Metrics (e.g. MAE, MSLE).
(Optional) Set a Threshold Type with upper and lower bounds for each metric. If left blank, the system will automatically configure thresholds based on your reference data.

Step 4: Univariate Data Drift

Enable this step to analyze features individually.

Figure 4. Univariate Data Drift Configuration

Select the Features you wish to analyze.
(Optional) Assign a Type per feature (Numerical or Categorical). If unassigned, the system defaults to the most appropriate one.
Assign a Method per type (Numerical or Categorical). If unassigned, the system uses the default one.
(Optional) Set a Threshold Type with upper and lower bounds for each method. If left blank, the system will automatically configure thresholds based on your reference data.

Step 5: Multivariate Data Drift (Optional)

Enable this step to analyze how features behave as a collective set.

Figure 5. Multivariate Data Drift Configuration

Select the Features for analysis.
(Optional) Set a Threshold Type with upper and lower bounds. If left blank, the system will automatically configure thresholds based on your reference data.

Step 6: Review and Finalise

Review your configurations and click Finalise to deploy the observability service for your project.

Review And Finalise - Cold Start

Figure 6. Review And Finalise - Cold Start

Review And Finalise - Chunk

Figure 7. Review And Finalise - Chunk

Review And Finalise - Performance Monitoring

Figure 8. Review And Finalise - Performance Monitoring

Review And Finalise - Univariate Data Drift

Figure 9. Review And Finalise - Univariate Data Drift

Review And Finalise - Multivariate Data Drift

Figure 10. Review And Finalise - Multivariate Data Drift

Prev
Get Familiar

Next
Model Monitoring Dashboard