Setting 'user_splits_fixed' in categorical binning about optbinning HOT 7 CLOSED

Hi @guillermo-navas-palencia
The splits result with default option of OptimalBinning is [[ 2., 7., 9., 3., 10., 4. 8],[-1]]. When I parse this splits as an option into user_splits parameter, it return the same error :).

from optbinning.

I created this example to reproduce your problem:

import numpy as np

np.random.seed(0)
n = 100000

x = sum([[i] * n for i in [-1, 2, 3, 4, 7, 8, 9, 10]], [])
y = list(np.random.binomial(1, 0.011665, n))
y += list(np.zeros(n))
y += list(np.random.binomial(1, 0.0133333, n))
y += list(np.random.binomial(1, 0.166667, n))
y += list(np.zeros(n))
y += list(np.random.binomial(1, 0.0246041, n))
y += list(np.zeros(n))
y += list(np.random.binomial(1, 0.025641, n))

user_splits = [[2., 7., 9., 3., 10., 4.], [8], [-1]]
user_splits_fixed = [True, True, True]

optb1 = OptimalBinning(dtype="categorical", user_splits=user_splits)
optb2 = OptimalBinning(dtype="categorical", user_splits=user_splits,
                       user_splits_fixed=user_splits_fixed)

for optb in (optb1, optb2):
    optb.fit(x, y)
    optb.binning_table.build()
    assert optb.binning_table.iv == approx(0.09345086993827473, rel=1e-6)

After commit a6d015b, it must work as expected.

from optbinning.

Hi @nic9lif3,

Thanks, I will try to reproduce your problem.

from optbinning.

Hi @nic9lif3,

There was a bug, I just fixed it, thanks for noticing. I acknowledge some tests are missing for the categorical type. I will commit the changes to master with new unit tests in a few minutes.

from optbinning.

Thanks @guillermo-navas-palencia for your contribute.

from optbinning.

Hi, @guillermo-navas-palencia ,

It seems like you change the code of setting user split for categorical. Error unhashable type list raises when I set user split is a list of list.

from optbinning.

Hi @nic9lif3,

No, it has not changed. I just run the previous example with version 0.8.0, and it worked without issues. It would be more helpful if you could attach the error message, my impression is that this error is not OptBinning related.

from optbinning.