Added lightweight ExtraBasicDiscretizer which assumes user knows which columns they want to discretize and wants one-hot output. This makes it really straightforward to generate meaningful column names at the end (e.g. balance_100_to_1000). Also, detecting which columns to discretize is kind of tricky in general so I wanted to have a class that avoids this
Fix bug where .dcols is destructively modified in existing discretizers
Fix the way existing discretizers check if a column is numeric — X[col].dtype in ['float', 'int'] doesn't always work
ExtraBasicDiscretizerwhich assumes user knows which columns they want to discretize and wants one-hot output. This makes it really straightforward to generate meaningful column names at the end (e.g. balance_100_to_1000). Also, detecting which columns to discretize is kind of tricky in general so I wanted to have a class that avoids this.dcolsis destructively modified in existing discretizersX[col].dtype in ['float', 'int']doesn't always work