search

PacktPublishing / Deep-Reinforcement-Learning-Hands-On

Hands-on Deep Reinforcement Learning, published by Packt
MIT License
2.83k stars 1.28k forks source link

Upgrade tensorboardX to 1.6 to avoid failure without cuda #24

Closed glasslion closed 5 years ago

glasslion commented 5 years ago

tensorboard-pytorch has been renamed to tensorboardX. The old version(0.7.1) would crash when I run Chapter03/03_atari_gan.py without cuda.

When calling writer.add_image("fake", vutils.make_grid(gen_output_v.data[:64], normalize=True), iter_no), it raises an error which says 'input tensor should be one of numpy.ndarray, torch.cuda.FloatTensor, torch.FloatTensor'. But without cuda, the tensor type is torch.Tensor.

The latest version of tensorboardX has fixed this issue.