pylab.plot "cannot convert float NaN to integer" after calling scipy.stats.multivariate_normal

Question

While testing a regression algorithm I found this strange behavior: for some covariance matrices, the multivariate_normal function gives correct samples but then an exception is raised (only) the first time pylab.plot() is called:

ValueError: cannot convert float NaN to integer

The following code reproduces the error:

import numpy as np
from scipy.stats import multivariate_normal as mnorm
from matplotlib import pyplot as plt

B = np.array([ 0, 0, 0])

# works fine
v1 = np.array([[1, 0, 0],
              [0, 1, 0],
              [0, 0, 1]])


# OK. non positive semidefinite, well raised exception
v2 = np.array([[ 0.2 , -0.2, -0.3],
              [-0.2,  0.4, -0.9],
              [-0.3, -0.9,  0.7]])

# KO. exception (?)
v3 = np.array([[ 0.2 , -0.02, -0.026],
              [-0.02,  0.014, -0.009],
              [-0.026, -0.009,  0.017]])



w = mnorm(mean=B, cov=v3).rvs()
print w

plt.plot(w)
plt.show()

And if plt.plot(w) is called a second time, then it works. Any ideas?

Versions:

python 2.7.5 Anaconda 1.9.1 (64-bit)

scipy 0.14.0

matplotlib 1.3.1

numpy 1.8.1

Louis Louis · Accepted Answer · 2014-05-28T08:49:28

Well, it works fine here, and says :

[-0.72849048  0.15439657  0.00146853]

and shows :

enter image description here

I use python 2.7.6

other packages are same as yours.

Hope it helped. Good luck !

pylab.plot "cannot convert float NaN to integer" after calling scipy.stats.multivariate_normal

1 Answers