Skip to main content

Task 5: Diagnose model

In this task, we will create a new model diagnostic based on the successfully completed experiment from task 4.

  1. In the H2O Driverless AI navigation menu, click DIAGNOSTICS.
  2. On the Model Diagnostics page, click +DIAGNOSE MODEL.
  3. For Diagnosed experiment, select the experiment tutorial-4a created in task 3.
  4. For Test dataset, select the loan_prediction_test.csv dataset.
  5. Click Launch diagnostics. Launch diagnostics
    note

    You may need to refresh the Model Diagnostics page (by clicking Diagnostics in the navigation menu again) to view the new model diagnostic.

  6. Click the new model diagnostic in the Model diagnostics table to open it.

The new model diagnostic page has the following informaiton:

diagnostic info

a. Scores: H2O Driverless AI calculates all available scores for the experiment. For the given model, it provides a comprehensive view of all possible scores relevant to a regression task.

Regression metric plots

Regression metric plots include the following graphs:

b. Actual vs predicted: compares the model's predicted values against the actual target values (loan amounts) from the dataset. This scatter plot helps assess the model's predictive accuracy by illustrating how closely the predictions align with the actual outcomes. The perfect model has a diagonal line.

c. Residual plot: displays residuals (the differences between actual and predicted values) against the predicted values. For a perfect model, residuals are zero.

d. Residual histogram: illustrates the distribution of residuals, providing insights into their spread and central tendency. A symmetric, bell-shaped distribution centered around zero suggests that the model's errors are normally distributed.

e. Download predictions: Click Download predictions to download the diagnostic predictions as a CSV file.

Congratulations, you have successfully created a new model diagnostic for your experiment. In Task 6, you will interpret the model by generating an MLI report.

Feedback