Neural Networks

Note: Neural networks are currently an experimental feature implemented in the nn-mt branch.

Neural networks are one of the models for signal set predictions (time-series forecasting) implemented in IVIS. Note that they are an optional feature and have to be enabled by the server admin.

To quickly create a new neural network prediction model, we provide a Neural Networks Quick Start guide. The available settings and parametrization of neural networks are described in more detail on the Add Model reference page.

The list of the implemented neural network architectures can be found in Neural Networks Architectures.

When the prediction is created, a Model overview page is available with the results of the training and options to enable and disable automatic predictions.

Enabling neural networks

This section is intended for IVIS instance admins with appropriate permissions and access to the server.

To enable the neural-network-based predictions, change the following in the yaml server config file (/server/config/production.yaml):

predictions:
  neural_network: true

After restarting the server, the Neural Network Training and Neural Network Prediction tasks should appear in the list of System tasks in settings, and they should successfully build.

The TensorFlow library used by the tasks requires python3-dev, which might not be installed by default. It can be installed using for example yum install python3-devel on CentOS or apt-get install python3-dev on Ubuntu.