segfault on gaussian_naive_bayes.apply() with SparseRealFeatures

[policecar]

No description provided.

[lisitsyn]

thanks, scheduled to fix ;)

[lisitsyn]

policecar,

could you please pull latest sources and check if it is failing still? I didn't managed to reproduce your issue, but fixed some errors that were in the classifier code.

Thanks.

[policecar]

i pulled, compiled, and ran the code again – no more segfault : )
but this time i get

Python(16510) malloc: *** error for object 0x100f8d208: incorrect checksum for freed object - object was probably modified after being freed. *** set a breakpoint in malloc_error_break to debug

[lisitsyn]

Could you please attach code reproducing it?

[policecar]

def gaussiannaivebayes_sparse():

import numpy as nu
from shogun.Features import SparseRealFeatures, Labels
from shogun.Classifier import GaussianNaiveBayes

data = nu.load('./tmp/feats.npz')
feats_train = data['feats_train'].item()
feats_test = data['feats_test'].item()
labels = data['labels']

feats_train = SparseRealFeatures( feats_train.astype(nu.float64).transpose() )
labels = Labels( nu.array( labels[0], dtype=nu.float64 ))
feats_test = SparseRealFeatures( feats_test.astype(nu.float64).transpose() )

gnb=GaussianNaiveBayes( feats_train, labels )
gnb_train = gnb.train()
output=gnb.apply( feats_test ).get_labels()
return gnb, gnb_train, output

if __name__=='__main__':
print 'GaussianNaiveBayes'
gaussiannaivebayes_sparse()

there's still an error in my test data but the malloc_error occurs equally if i apply naive bayes to the training data

[policecar]

looks like my current code runs without throwing any errors...
thanks for the quick fix the other day!

[lisitsyn]

I'm sorry I didn't react on last message. Anyway, I'll close this issue after one more fix as it is working now

[policecar]

one more thing, how about computing the conditional probabilities using the log which would yield a sum instead of the product of all conditional probabilities per feature vector, and thereby avoiding issues with underflow ?

[policecar]

hehe, i just noticed that's what you just did. thanks plenty!

[lisitsyn]

Well that's exactly what I did in a14a48a :)