Neural Networks Model Overview - smartarch/ivis-core GitHub Wiki

When a new model is created, the user is automatically redirected to the model overview page. This page can also be accessed from the predictions table for the signal set.

Tools

The buttons at the top of the page allow the user to perform several actions with the model.

Re-run training will delete the trained model and run the training and tuning process again.

When automatic predictions are enabled, the prediction job runs as soon as new data arrive to the signal set. The Enable/Disable automatic predictions button can be used to enable or disable this behaviour.

Latest predictions

After selecting signals, line charts with historical data (in black) and the predictions will appear. A signal for each aggregation function and the combined signal (as described here) are available for selection. The minimum aggregation is drawn as a blue line, the average aggregation is drawn as a purple line and the maximum aggregation is drawn as a red line. The combined signals is drawn as a purple area. Each signal from the original signal set will be displayed in a separate line chart.

The following image shows a line chart for the combined signal and two aggregation signals: minimum and maximum.

Training results

Best found hyperparameters

This sections displays a table with the values of the best found hyperparameters. These are the parameters of the saved model which is used for generating the predictions.

Trials

The trials table shows the loss for each configuration of hyperparameters tested during tuning.

Training log

The last section of the page displays the log of the training job. The log can be used to see the progress of the ongoing tuning and training.

The run logs are periodically deleted based on the server configuration, so the log might not be available when the overview page is accessed after the training is finished.