Module view.NetworkTrainer
Widget that allows the user to define the layers of a neural network.
Functions
NetworkTrainer.new (ctrl, parent) | Constructor. |
NetworkTrainer:clear () | Clears the training status (interface only, doesn't changed the network) |
NetworkTrainer:run (project) | Runs the dialog. |
NetworkTrainer:training_callback (max_epochs, rep_epochs, target, cur_epoch) | Training callback. |
NetworkTrainer:update_status (epoch, epochs) | Updates the training status, based on parameters of the current network. |
(private) NetworkTrainer:add_status_ui () | |
(private) NetworkTrainer:add_stop_conditions_ui () | |
(private) NetworkTrainer:build_gui (parent) | Builds the interface. |
(private) NetworkTrainer:clear_reports () | Clears the stored report information (MSE, weights). |
(private) NetworkTrainer:create_plot_dialog () | |
(private) NetworkTrainer:prepare (project) | Prepares the dialog. |
(private) NetworkTrainer:queue_update () | Queues a GUI update. |
(private) NetworkTrainer:show_bit_dialog () | |
(private) NetworkTrainer:show_mse_dialog () | |
(private) NetworkTrainer:show_weight_dialog () | |
(private) NetworkTrainer:start_training () | Starts the training. |
(private) NetworkTrainer:stop_training () | Stops the training (queue it, the training only stops on the next iteration). |
(private) calculate_interval (net, data) | Estimates a good epochs_between_reports param, based on the size of the training data and the number of connections in the network. |
Functions
- NetworkTrainer.new (ctrl, parent)
-
Constructor.
Parameters
- ctrl: Application controller, used to emit events
- parent: Parent window
Return value:
New instance of NetworkTrainer - NetworkTrainer:clear ()
- Clears the training status (interface only, doesn't changed the network)
- NetworkTrainer:run (project)
-
Runs the dialog.
Parameters
- project: Project to use the network and training data
- NetworkTrainer:training_callback (max_epochs, rep_epochs, target, cur_epoch)
-
Training callback. Since the training occurs in the same thread as the interface, we process the pending main loop events here.
Parameters
- max_epochs:
- rep_epochs:
- target:
- cur_epoch:
- NetworkTrainer:update_status (epoch, epochs)
-
Updates the training status, based on parameters of the current network.
Parameters
- epoch:
- epochs: Current epoch (optional)
- (private) NetworkTrainer:add_status_ui ()
- (private) NetworkTrainer:add_stop_conditions_ui ()
- (private) NetworkTrainer:build_gui (parent)
-
Builds the interface.
Parameters
- parent: Parent window
- (private) NetworkTrainer:clear_reports ()
- Clears the stored report information (MSE, weights).
- (private) NetworkTrainer:create_plot_dialog ()
- (private) NetworkTrainer:prepare (project)
-
Prepares the dialog.
Parameters
- project:
- (private) NetworkTrainer:queue_update ()
- Queues a GUI update. Used to avoid the useless concentration of events in the same callback.
- (private) NetworkTrainer:show_bit_dialog ()
- (private) NetworkTrainer:show_mse_dialog ()
- (private) NetworkTrainer:show_weight_dialog ()
- (private) NetworkTrainer:start_training ()
- Starts the training.
- (private) NetworkTrainer:stop_training ()
- Stops the training (queue it, the training only stops on the next iteration).
- (private) calculate_interval (net, data)
-
Estimates a good epochs_between_reports param, based on the size of the training data and the number of connections in the network.
Parameters
- net: Network to be used in the training
- data: Data to be used in the training
Return value:
Report interval, an integer between 1 and 1000