Closed
Description
I propose to refactor the input validation. The current zoo of methods is kinda confusing.
Related to #3142.
Checks that we want are:
- numpy array vs sparse vs list vs anything indexable
- sparse matrix type
- inf / nan
- dtype
- ndims
- number of samples is consistent in multiple arrays.
- contiguous
I'll now check if there is anything else that we currently check and see what functions we have.
Remaining Issues
- possibly remove "make_float_array" and "1d_or_column" and also make these options of check_array.
- Currently "ensure2d" makes vectors into rows, which I find pretty counter-intuitive. This is for backward compatibility. 1d input is not currently handled consistently.