You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There are 2 methods for handling categorical features. By default, all categoricals are converted to integers. There is no limit for a number of categories. However, there is also an additional step during AutoML training called mix_encoding. In this step, the categorical features with less than 21 categories are treated with binary one-hot encoding. The mix_encoding is available by default in the Compete step. If you want to enable/disable mix_encoding please set it in AutoML() constructor.
In the above example, the AutoML will tune algorithms with Optuna and will also try Xgboost with mix_encoding (integer + one-hot encodings).
Important notes
mix_encoding will not be used (even if set in constructor) if all categorical features have more than 20 categories. There needs to be at least one feature with less than 21 categories.
mix_encoding is used only for the Xgboost algorithm (probably can add some enhancement here ...)
In your blog, I remembered that you mentioned the largest number of categorical features was 25. Is it true? My dataset has 85 categorical columns.
The text was updated successfully, but these errors were encountered: