GaussianNB :- ValueError: The sum of the priors should be 1 - python-2.7

Question

What i am trying to do?
I am trying to train the dataset which has 10 labels using GaussianNB classifier but while tunning my gaussianNB prior parameters i am getting this error:-
File "/home/mg/anaconda2/lib/python2.7/site-packages/sklearn/naive_bayes.py", line 367, in _partial_fit
raise ValueError('The sum of the priors should be 1.')
ValueError: The sum of the priors should be 1.
Code for this:-
clf = GaussianNB(priors = [0.08, 0.14, 0.03, 0.16, 0.11, 0.16, 0.07, 0.14, 0.11, 0.0])
You can see the sum is clearly 1 but it showing me this error, can you point the error.

Answer 1

This looks like a pretty bad design-decision within sklearn as they are doing the usual don't compare floating-point numbers stuff (what every computer scientist should know about floating-point arithmetic), which surprises me (as sklearn is usually high-quality code)!
(I don't see any wrong usage on your end, despite using a list. The docs call for an array, not array-like like in many other cases, but their code is doing the array-conversion nonetheless)
Their code:
if self.priors is not None:
priors = np.asarray(self.priors)
# Check that the provide prior match the number of classes
if len(priors) != n_classes:
raise ValueError('Number of priors must match number of'
' classes.')
# Check that the sum is 1
if priors.sum() != 1.0:
raise ValueError('The sum of the priors should be 1.')
# Check that the prior are non-negative
if (priors < 0).any():
raise ValueError('Priors must be non-negative.')
self.class_prior_ = priors
else:
# Initialize the priors to zeros for each class
self.class_prior_ = np.zeros(len(self.classes_),
dtype=np.float64)
So:
You give a list, but their code will create an numpy-array
Therefore np.sum() will be used for summing
There might be fp-math related numerical-errors in summing like in your case
your sum is technically != 1.0; but very close to it!
fp-comparison x == 1.0 is considered bad!
numpy brings np.isclose() which is the usual approach of doing this
Demo:
import numpy as np
priors = np.array([0.08, 0.14, 0.03, 0.16, 0.11, 0.16, 0.07, 0.14, 0.11, 0.0])
my_sum = np.sum(priors)
print('my_sum: ', my_sum)
print('naive: ', my_sum == 1.0)
print('safe: ', np.isclose(my_sum, 1.0))
Output:
('my_sum: ', 1.0000000000000002)
('naive: ', False)
('safe: ', True)
Edit:
As i think that this code is not good, i posted an issue here which you can follow to see if they comply or not.
numpy.random.sample(), which also takes such a vector, is actually doing a fp-safe approach too (numerically more stable summation + epsilon-check; but not using np.isclose()) as seen here.