Skip to content

Follow-up the implementation #7

New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Merged
merged 5 commits into from
May 23, 2019
Merged

Conversation

fperdigon
Copy link
Contributor

@fperdigon fperdigon commented May 22, 2019

  • label_guessin function added
  • to_categorical function added
  • Modify the MixUp according to the MixMatch paper
  • Fix some small error
  • Adding some comments
  • Specify in the README.md that the implementation is not finished

fperdigon added 5 commits May 22, 2019 16:23
The MixUp used the paper is a slightly modified version to avoid problems with the SSL loss.
Also, the name of the function was changed to mixup_mod to be explicit that is a modified version.
Adding Label Guessing function
This function converts the NumPy array in Pytorch tensors. Then predict the labels. 
The predictions average is returned in NumPy array type, this helps to keep all variables inside MixMatch in NumPy type.
The function was added in the mixmatch function.
Normally the PyTorch labels for classification are in integer numbers associated with classes.
For the MixMatch algorithm, the labels need to be in one hot format.
The to_categorical function was taken from Keras code, is able to handle batches.
num_clases=10 for future CIFAR test.
Inside of the algorithm y is converted to one hot encode format. 
At the end of the MixMatch phase, the labels p and q are converted back to the natural classes format in PyTorch
- Fix the indices generation
- Wy must be the concatenation of Uy and y (before qb and y, this generated a size mismatch with Ux)
- Adding some comments
Adding that the project is not finished in the README.md
@gan3sh500 gan3sh500 merged commit b23ce98 into gan3sh500:master May 23, 2019
# for free to join this conversation on GitHub. Already have an account? # to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants