This is the default location of results from training a code. Using the configuration files provided in the conf directory, results will be saved according to the following directory hierarchy:
<dataset>/<base model>/k<parameter k>/<loss function>/<encoder architecture>
The following files will be saved:
{train, val}_loss.txt: Average loss for each epoch.{train, val}_reconstruction_accuracy.txt: Reconstruction accuracy attained in each epoch of training. Recall from the paper that reconstruciton-accuracy is defined as the accuracy of decoded reconstructions with respect to the classifications of the underlying base model.{train, val}_overall_accuracy.txt: Overall-accuracy attained in each epoch of training. Recall from the paper that overall-accuracy is defined as the accuracy of decoded reconstructions with respect to the true labels associated with a particular dataset.current.pth: PyTorch checkpoint containing the latest snapshots of the encoder and decoder parameters, encoder and decoder optimizers, and current epoch.best.pth: The state dict among all epoch so far that has achieved the highest reconstruction-accuracy on the validation dataset.