dnouri / gnumpy Goto Github PK

mltest.docx
Hello,
I'm trying to run a deep learning algorithm. I followed the code from Adrian's tutorial.
I'm getting the following error:

File "mltest.py", line 4, in
from nolearn.dbn import DBN
File "C:\Users\User\AppData\Local\Programs\Python\Python35-32\lib\site-packages\nolearn\dbn.py", line 5, in
from gdbn.dbn import buildDBN
File "C:\Users\User\AppData\Local\Programs\Python\Python35-32\lib\site-packages\gdbn\dbn.py", line 25, in
import gnumpy as gnp
File "C:\Users\User\AppData\Local\Programs\Python\Python35-32\lib\site-packages\gnumpy.py", line 225
except _cudamat.CUDAMat Exception as e: # this means that malloc failed
^
SyntaxError: invalid syntax

I'm running Python 3.5.2, OpenCV 3.0, corresponding nolearn and scikit modules on Windows 10 Pro.
Can someone help me troubleshoot this issue?

Thank you,

File "/opt/conda/lib/python3.6/site-packages/gnumpy-0.2-py3.6.egg/gnumpy.py", line 52
except: print 'gnumpy: failed to import cudamat. Using npmat instead. No GPU will be used.'; _useGpu = 'no'
^
SyntaxError: Missing parentheses in call to 'print'. Did you mean print('gnumpy: failed to import cudamat. Using npmat instead. No GPU will be used.')?

import numpy as np
import gnumpy as gp
gp.garray(np.ones([10000,100000]))
garray([[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
...,
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.],
[ 1., 1., 1., ..., 1., 1., 1.]])
gp.garray(np.ones([100000,100000]))
Traceback (most recent call last):
File "", line 1, in
File "/users4/qkshi/anaconda2/lib/python2.7/site-packages/gnumpy.py", line 731, in init
if npa.size!=0: cm.numpy_array[:] = npa.reshape((-1, 1), order='C') # no cudamat.reformat is needed, because that's only dtype and order change, which are handled by the assignment anyway
ValueError: could not broadcast input array from shape (10000000000,1) into shape (1410065408,1)

when indexing a gnumpy.garray using a tuple of 2 lists of indices, gnumpy's implementation of __getitem__ method casts the lists to the native gpu data type which is 32 bit float. Therefore, due to floating point round errors, indexing is not correct for moderately large matrices.
Bug can be reproduced by the following code:

import gnumpy as gnp
a = gnp.randn(5000,5000)
if a.as_numpy_array()[([4001],[4001])] != a[([4001],[4001])]:
    print 'bug!'