niboshi

Add an attribute to FunctionNode to indicate it supports arrays on different devices

2

This idea came from an offline discussion related to #454. Currently only array types are checked in FunctionNode inputs, but not their devices. If the arrays are on different GPU...

push: Resolve "from" commit

1

Fixes #16 It looks for a commit that is an ancestor of `HEAD` and included in any branch in the `origin`.

I found that untracked (in terms of git) files are not synchronized. Is this by design? This can be a trap if you don't understand, and I think it's rather...

Segmentation fault in combination with certain libraries

3

```py import _hashlib import mysql.connector import pfio dest_path = 'hdfs:///my/hdfs/file' with pfio.open(dest_path, 'wb') as file_out: file_out.write(b'data') ``` In my environment, the above code results in a segmentation fault: ``` python:...

#### Code ```py import cupy f = cupy.ElementwiseKernel( 'T a, U b', 'T out', ''' out = b + a; ''') x = cupy.ones((2, 3), 'float32') y = cupy.float32(2) f(x,...

Policy: performance vs compatibility

1

Context: https://github.com/cupy/cupy/pull/2483#discussion_r328074119 There are cases where we have to make a choice between performance vs compatibility. In the case above, `scipy.ndimage.convolve` (`cupyx.scipy.ndimage.convolve`) internally casts the weights to `float64` for better...

The following kind of information would be helpful to investigate test results if they were in the logs. * Python version * Environment variables

Regarding the issue https://github.com/chainer/chainer/issues/3878. Currently we don't have test with CHAINER_DEBUG enabled, but we should test it as well. It may be fine to enable it all the time, but...

Rel: #512 If tests are run from within the chainer directory, `chainer` package installed by `pip install ./` won't be imported; instead, source files existing in the directory are imported....

https://github.com/chainer/chainer-pytorch-migration/pull/8#issuecomment-576228210