search

zuoxingdong / VIN_PyTorch_Visdom

PyTorch implementation of Value Iteration Networks (VIN): Clean, Simple and Modular. Visualization in Visdom.
226 stars 39 forks source link

The 28x28 gridworld file (dataset for training) corrupted? #2

Closed kentsommer closed 7 years ago

kentsommer commented 7 years ago

I receive the following error when trying to train on the 28x28 gridworld dataset:

cPickle.UnpicklingError: BINUNICODE pickle has negative byte count

I have tried re-downloading (which did not fix it), and have confirmed both the 16x16 and 8x8 versions work fine.

avivt commented 7 years ago

OK, let me check it out.

On Tue, Apr 18, 2017 at 12:34 AM, Kent Sommer notifications@github.com wrote:

I receive the following error when trying to train on the 28x28 gridworld dataset:

cPickle.UnpicklingError: BINUNICODE pickle has negative byte count

I have tried re-downloading (which did not fix it), and have confirmed both the 16x16 and 8x8 versions work fine.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/zuoxingdong/VIN_PyTorch_Visdom/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AOeQNeSkFDqTePgsfSsjQt3iGQNnvptmks5rxGd5gaJpZM4M_-jT .

kentsommer commented 7 years ago

@avivt

I was able to generate my own dataset. So, my issue is fixed. However, I'll leave this open as I'm sure at some point someone would like to use the included 28x28 and a fix here would be useful.

zuoxingdong commented 7 years ago

@kentsommer Sorry for the late reply. I've tried again to re-clone whole repo to test it, however, I am not able to reproduce this error. Are you using Anaconda, and with Python 3.x ? Maybe it is also related to your pickle version.

kentsommer commented 7 years ago

@zuoxingdong

No, I'm was virtualenv and Python 2.7 which after trying 3.x seems to fix the issue... Odd that the only issue with 2.7 and > is the loading issue on 28x28. Regardless this is working if using Python 3.x, closing out.