CODEWITHSUNDEEP
pythonscikit-learnscipy
fcol
fcol

Reputation: 189

Save (pickle) Scipy KDE

How can I pickle or save a scipy kde for later use? 

import scipy.stats as scs
from sklearn.externals import joblib

kde = scs.gaussian_kde(data, bw_method=.15)
joblib.dump(kde, 'test.pkl')

I tried above and received this error:

PicklingError: Can't pickle <function gaussian_kde.set_bandwidth.<locals>.<lambda> at 0x1a5b6fb7b8>: it's not found as scipy.stats.kde.gaussian_kde.set_bandwidth.<locals>.<lambda>

Upvotes: 3

Views: 1749

Answers (1)

Kevin
Kevin

Reputation: 8207

Looks like joblib is having trouble with the set_bandwith method, my guess is because of the lambda function in the method -- pickling lambdas has been discussed here.

with open('test.pkl', 'wb') as fo:  
    joblib.dump(lambda x,y: x+y, fo)

PicklingError: Can't pickle <function <lambda> at 0x7ff89495d598>: it's not found as __main__.<lambda>

cloudpickle and dill both work as far as I can tell:

import cloudpickle
import dill

with open('test.cp.pkl', 'wb') as f:
    cloudpickle.dump(kde, f)  

with open('test.dill.pkl', 'wb') as f:
    dill.dump(kde, f)

with open('test.cp.pkl', 'rb') as f:
    kde_cp = cloudpickle.load(f)

with open('test.dill.pkl', 'rb') as f:
    kde_dill = dill.load(f)

Inspect some of the data:

import numpy as np

print(np.array_equal(kde.dataset, kde_cp.dataset))
True
print(np.array_equal(kde.dataset, kde_dill.dataset))
True
print(np.array_equal(kde_cp.dataset, kde_dill.dataset))
True

kde.pdf(10) == kde_cp.pdf(10) == kde_dill.pdf(10)
array([ True])

Upvotes: 5

Related Questions