Addresses an issue raised in #43 and adds a new feature:
Makes the default for parallel equal to the None-ness of resume_path. The reason for this is that if resume_path is None, then this is likely a new model to be fed into train, which expects a model that is not wrapped in DataParallel
Allows for custom accuracy function to be supplied via the custom_accuracy training argument. The function should return a tuple (top1, top5) given the arguments (out, target).
Addresses an issue raised in #43 and adds a new feature:
Makes the default for
parallelequal to the None-ness ofresume_path. The reason for this is that ifresume_path is None, then this is likely a new model to be fed intotrain, which expects a model that is not wrapped in DataParallelAllows for custom accuracy function to be supplied via the
custom_accuracytraining argument. The function should return a tuple(top1, top5)given the arguments(out, target).Update docs for example usage of custom_accuracy.