GaussianProcess with batch_size gives "'float' object cannot be interpreted as an integer"

[stmax82]

Windows-10-10.0.10586
Python 3.4.4 |Continuum Analytics, Inc.| (default, Jan 29 2016, 15:20:20) [MSC v.1600 64 bit (AMD64)]
NumPy 1.10.4
SciPy 0.17.0
Scikit-Learn 0.17.1

The following code:

import numpy as np
from sklearn.gaussian_process import GaussianProcess

X = np.random.rand(100, 3)
y = np.random.rand(100)

gp = GaussianProcess().fit(X, y)

gp.predict(X, batch_size=5)

Results in the error

'float' object cannot be interpreted as an integer

Because of:

C:\...\lib\site-packages\sklearn\gaussian_process\gaussian_process.py in predict(self, X, eval_MSE, batch_size)
    520 
    521                 y = np.zeros(n_eval)
--> 522                 for k in range(max(1, n_eval / batch_size)):
    523                     batch_from = k * batch_size
    524                     batch_to = min([(k + 1) * batch_size + 1, n_eval + 1])

TypeError: 'float' object cannot be interpreted as an integer

[glouppe]

GaussianProcess has been deprecated in 0.18. If this is possible for you, please update to the dev version, where will now find GaussianProcessRegressor instead.

[stmax82]

oh thanks, I didn't know the GaussianProcess class was deprecated (there's no warning anywhere). Wish I could try the new one in 0.18, but I cannot use dev versions in production.

Will there be a 0.17.2 release before 0.18? The fix would be trivial - I think replacing the division "/" by "//" should do it..

[nelson-liu]

@glouppe according to the docstring at https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/gaussian_process/gaussian_process.py#L61, it will be deprecated in 0.20. Is this an error in the docstring?

[glouppe]

No, this is correct. The new GP implementation has been added after the 0.17 release, and therefore the old implementation will be deprecated in 0.20, two release cycles after the next 0.18 release.

GaussianProcess is deprecated starting from 0.18 but it is still there until 0.20, where it will be entirely removied.

[nelson-liu]

ah that makes sense, thanks for the info.