Path3.Mod1.f - Automated Machine Learning - Evaluate and Compare Models In ML Studio

Question 1

DA BM, EM VE

• ML Studio > AutoML Experiment > Overview Page - Two things this page shows you
• ML Studio > AutoML Experiment > Models Page - What you can Explore and what you can View…

Answer 1

• Review input Data Asset and summary of the Best Model
• Explore Models that were trained (by algorithm names), with View Explanation for the best one, and Responsible AI Dashboard

Question 2

CBD MFVI HCFD

Data Guardrails:
- Where they are located
- Three data guardrails auto-applied to classification models

Answer 2

Job Details > Data Guardrails tab. A training job has to complete first before you can view what guardrails were applied.

• Class Balancing Detection - Imbalanced or underrepresented classes
• Missing Feature Value Imputation - Provide values when missing (average, most common value, etc.)
• High Cardinality Feature Detection - Reduce fields that appear to have high cardinality (nearly or always a random value)

Question 3

Data Guardrails are applied when…

Answer 3

…Featurization is enabled for your experiment.

Question 4

P D A

Data Guardrails have three possible states

Answer 4

• Passed: No problems, no action required
• Done: Changes applied to your data, review changes made by AutoML
• Alerted: An issue was detected and couldn’t be fixed. Review data to fix

Question 5

Where AutoML displays the Scaling and Normalization techniques applied during training and in what format

Answer 5

Models Tab > Algorithm Name column. If a technique is applied, it is listed in this format: <applied scaling and/or normalization techniques,...>, <algorithm name>

Question 6

You can View Explanation for any trained model by selecting one of the trained models in the Overview tab and select the Explain Model subtab (T/F)

Answer 6

True. The explanation is an approximation of the model’s interprerability. This doesn’t go just for the Best Model/Top Pick. You can opt to see an Explanation and even Responsible AI for any Model per request.