Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Fallbacks of ChainerX on GetItem fail when indices contains chainerx.ndarray

See original GitHub issue

When chainerx with native backend falls back to numpy:

>>> chainer.Variable(chainerx.zeros(3, device='native'))[numpy.arange(3)]
variable([0., 0., 0.])
>>> chainer.Variable(chainerx.zeros(3, device='native'))[chainerx.arange(3, device='native')]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/functions/array/get_item.py", line 143, in get_item
    return GetItem(slices).apply((x,))[0]
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/function_node.py", line 315, in apply
    outputs = self.forward(in_data)
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/functions/array/get_item.py", line 41, in forward
    return utils.force_array(xs[0][self.slices]),
IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices

When chainerx with cuda:0 backend falls back to cupy:

>>> chainer.Variable(chainerx.zeros(3, device='cuda:0'))[cupy.arange(3)]
variable([0., 0., 0.])
>>> chainer.Variable(chainerx.zeros(3, device='cuda:0'))[chainerx.arange(3, device='cuda:0')]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/functions/array/get_item.py", line 143, in get_item
    return GetItem(slices).apply((x,))[0]
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/function_node.py", line 315, in apply
    outputs = self.forward(in_data)
  File "/home/fujita/.local/share/virtualenvs/chainerrl_20180827-9TPAa4mI/lib/python3.5/site-packages/chainer/functions/array/get_item.py", line 41, in forward
    return utils.force_array(xs[0][self.slices]),
  File "cupy/core/core.pyx", line 1650, in cupy.core.core.ndarray.__getitem__
  File "cupy/core/core.pyx", line 3255, in cupy.core.core._simple_getitem
TypeError: Invalid index type: <class 'chainerx.ndarray'>

I want these to work so I can write numpy/cupy/chainerx-agnostic code.

Issue Analytics

State:
Created 5 years ago
Comments:7 (7 by maintainers)

Top GitHub Comments

1reaction

beam2dcommented, Apr 15, 2019

It looks https://github.com/chainer/chainer/pull/6769 contains the fallback fix for this issue.

0reactions

muupancommented, Apr 23, 2019

Thank you so much for fixing the issue!

Top Results From Across the Web

chainer/_fallback_workarounds.py at master · chainer/chainer · GitHub

This file defines workaround implementation for. # NumPy-compatibility functions that fall back to NumPy/CuPy functions.

ChainerX Tutorial — Chainer 7.8.1 documentation

Graph construction and backpropagation is built into the array, meaning that any function, including the NumPy-like functions, can be backpropagated through. In ...

Chainer Documentation - Read the Docs

It contains a sorted indices of the input variables w.r.t. which the gradients are ... held by the Variable object is not a...

Chainer and ChainerX - Nvidia

W2.array -= lr * W2.grad. Write forward prop as a plain Python script. Variables hold how they were computed. Use it to compute...

A flexible framework of neural networks for deep learning

import numpy as np import cupy import chainer import chainerx as chx ... of ChainerX on GetItem fail when indices contains chainerx.ndarray ......