Memory leak with conversion from numpy array to System.Single
I have been using Python for .Net quite successfully to access an I/O library that allows me to perform some numerical calculations in Numpy. This library primarily requires the data to be fed back as Single Arrays. However when I perform many conversions from numpy arrays (or even just lists) to Single I am experiencing a memory leak that eventually causes a crash.
The following distills the problem. By switching between the two different Array types (Single & Double) the leak is controlled so it would seem there is some issues with Single. It does not appear to matter whether change the dtype of the numpy array to float32. I have also tried Array.ConvertAll but the same problem arrises.
import numpy as np import clr from System import Array, Single, Double
N = 100000 M = 1000
for i in range(N):
# Numpy calculations A = np.random.random( M )
# Convert to single array - creates memory leak sgl_arry = Array[Single](A)
# Convert to double array - does NOT create memory leak
#dbl_arry = Array[Double](A)