CIFAR10-img-classification-tensorflow
CIFAR10-img-classification-tensorflow copied to clipboard
deep-diver
Low accuracy in notebook and py
For some reason I'm hovering around ~0.15 accuracy during the training for both the notebook and the py file. Also, in the final layer, shouldn't it be full4 not full3? To note, I am using python 2.7, so have had to slightly change the import for urlretrieve and also remove the extra param of "latin1" from the pickle opens. Could this have any affect? Further more the py can't find the "helper" module (no easy fix even when copying the other functions in the notebook in). Thanks.
If you have any idea why my accuracy is so low in training any response would be appreacited.
Did you ever get to the bottom of this? I'm currently having the same issue.
I did but it was so long ago I've forgotten I'm afraid. If i remember correctly I think it wasn't a big problem, and just a small bug/change somewhere. I'm sorry I can't be more helpful.
On Wed, 15 Jul 2020 at 17:48, shooker531 [email protected] wrote:
Did you ever get to the bottom of this? I'm currently having the same issue.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/deep-diver/CIFAR10-img-classification-tensorflow/issues/7#issuecomment-658876607, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFYSMI5H2K6OZT4GUG72AKLR3XMWVANCNFSM4GWNP6QA .
I think I've found the issue. The cost should be tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=model, labels=y)) rather than tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y)). That seems to have got the accuracy to a good level for me.