Summary
In this chapter, we explored a powerful XAI tool. We saw how to analyze the features of our training and testing datasets before running an ML model.
We saw that Facets Overview could detect features that bring the accuracy of our model down because of missing data and too many records containing zeros. You can then correct the datasets and rerun Facets Overview.
In this iterative process, Facets Overview might confirm that you have no missing data but that the data distributions of one or more features have high levels of non-uniformity. You might want to go back and investigate the values of these features in your datasets. You can then either improve them or replace them with more stable features.
Once again, you can rerun Facets Overview and check the distribution distance between your training and testing datasets. If the Kullback-Leibler divergence is too significant, for example, you know that your ML model will produce many errors.
...
