Adding metrics to Callback functions + Fixing model summary issue
Summary
This MR introduces the following changes
- Fixing small issue with the model summary of the
DenseNet
. - Removing history object from training.
- Training metrics are now saved via the
MyCallback
functions in a separate file. - Two dicts are now written to disk when training: The
train_metrics_dict.json
and the normalvalidation.json
with the parameters used for validation in the name of the file. Thetrain_metrics_dict.json
contains the train metrics, like accuracy, loss, learning rate, while thevalidation.json
contains the same values (except learning rate) but for the validation dataset. - If no validation dataset is given, only the
train_metrics_dict.json
is produced. - Updating the
get_validation_file_name
function toget_metrics_file_name
. Outputs now the path to thetrain_metrics_dict.json
and the path + name for thevalidation.json
. - Unit tests are adapted.
Relates to the following issues
- Closes #173 (closed)