ENH: quantile vs equal interval bins about alphalens HOT 6 CLOSED

That makes sense to me. Have you thought about how the API would look like?

from alphalens.

My idea would be to add a flag to create_factor_tear_sheet so that the user can supply the binning type (quantile or equal interval). Other than that we don't need additional api changes but we need a function similar to performance.quantize_factor that accepts the same input but returns the factor binned by value intervals instead of quantiles. We then call performance.quantize_factor or the new function depending on the binning type flag.

from alphalens.

Yes, a kwarg for the binning type makes sense.

from alphalens.

I thought about the api change again and an option would be to remove the "quantiles=5" argument from create_factor_tear_sheet function and add this new argument:

"binning_mode=(mode, mode_args)"

E.g.:
binning_mode=("quantiles", 5) #Number of uniform quantiles
binning_mode=("quantiles", [0, .15, .5, .85, 1.] ) # list of non uniform quantiles
binning_mode=("bins", 5) #number of equal-width bins
binning_mode=("bins", [-1, -0.3, 0., 0.3, 1] ) # list of bin edges allowing for non-uniform bin width

Also there is room for future addition of new modes.

Note: when "quantiles" mode is selected we'll end up calling pandas.qcut and when "bins" is provided we'll call pandas.cut

from alphalens.

That looks a bit bloated to me. What if we just have quantiles and bins and only one of them can be not-None?

from alphalens.

That sounds good too, let's go for that then.

from alphalens.