This PR implements Tensorboard callback allowing to visualize graphs of the training and test metrics, as well as activation histograms for the different layers in the model.
PR introduces a Tensorboard class with custom set_model, on_epoch_end and on_train_end methods.
Currently monitoring:
Scalar variable summaries for training and validation loss, validation ROC, per epoch
Weights, biases and activations histograms and distributions, per each layers, per epoch
Graph visualization
Note: completely independent on Keras, but consider inheriting from Callback in future. Reason: current Keras implementation does not allow passing validation data as a generator, while loading the whole list causes OOM.
Note: free parameter is the number of validation steps, which is passed along with the validation data generator. Consider changing the value as appropriate
This PR implements Tensorboard callback allowing to visualize graphs of the training and test metrics, as well as activation histograms for the different layers in the model.
PR introduces a
Tensorboard
class with customset_model
,on_epoch_end
andon_train_end
methods.Currently monitoring:
Note: completely independent on Keras, but consider inheriting from Callback in future. Reason: current Keras implementation does not allow passing validation data as a generator, while loading the whole list causes OOM.
Note: free parameter is the number of validation steps, which is passed along with the validation data generator. Consider changing the value as appropriate