vjache commented on issue #17143: Wrong sum of array URL: https://github.com/apache/incubator-mxnet/issues/17143#issuecomment-568257313 > Hi @vjache , it is not a bug. > The reason is that the default data type is `float64` in numpy, but `float32` in MXNet. > > ```python > >>> import numpy as np > >>> import mxnet.ndarray as md > >>> a = [1.0504983e+07, 1.0000000e+00, 2.0000000e+00, 3.0000000e+00, 5.0000000e-01, 1.0000000e+01, > ... 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 0.0000000e+00, 0.0000000e+00] > >>> np.array(a).dtype, np.array(a).sum() > (dtype('float64'), 10504999.5) > >>> md.array(a).dtype, md.array(a).sum() > (<class 'numpy.float32'>, > [10505000.] > <NDArray 1 @cpu(0)>) > >>> np.array(a, dtype=np.float32).sum() > 10505000.0 > >>> md.array(a, dtype=np.float64).sum() > > [10504999.5] > <NDArray 1 @cpu(0)> > ``` Sorry, I've seen your response only after I posted my where I foun the bypass. But still I consider this is a serious thing. If I create an array and the accuracy is lost then I should be at least warned, and at best the error raised. It is incorrect to silently do decreasing of accuracy when I create array.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
