Incorporate Categorical Variables in Interactions

[LyndonFan]

Missing functionality

I often encounter product data, which are grouped and categorised in different ways. Having found the library recently and wanting to use it, I would like to quickly see how these categories interact with each other.

I think this would be useful for datasets which can be grouped in different ways. e.g. For the titanic dataset, we can look at the gender of passengers and whether they survived. Another example is for a dataset of products, we may want to look at how the distribution of sales varies for products in different categories.

Proposed feature

In the interaction section of the report, we can allow categorical variables to be chosen alongside continous variables, with new types of plots as such:

• categorical vs categorical: heatmap of frequencies
• categorical vs continuous: boxplot of distribution of the continuous variable in different categories

Alternatives considered

The correlations do provide some indication about how the categories overlap but it does not always suffice.

Additional context

I am happy to work on the implementation of this feature.

[fabclmnt]

Hi @LyndonFan, definitely this something missing. We would be happy to have your contribution with the development of this feature.

[aquemy]

Hi @LyndonFan,

This would be very useful. However, note that with high cardinality variables, it might be difficult to visualize a heatmap of frequency. How would you solve this problem?

Would you consider something like Weight of Evidence and Information Value for instance? Low and high WoE and IV could be displayed as it is mostly what we are interested in I suppose.