API Reference

In this section, implemented classes are explained and their methods are described. In addition, the utils functions implemented in this package are described. For these classes and functions, the necessary and optional parameters are documented here.

The package consists of two subpackages with different purposes. The models folder contains different architectures of neural networks implemented in PyTorch, including different neural networks classes for predicting regression tasks or time series forecasting. It also includes a wrapper to build ensembles of models, where the outputs of multiple models are averaged.

The second subpackage contains utils functions. There are functions for data creation and formatting, but also for evaluating and investigating your trained model, including an analysis of the hidden-layer correlation, input/output sensitivity and visualization.

Models

Implemented models for regression tasks are:

• Feed Forward Neural Network
• Deep Feed Forward Neural Network

If you have only a small number of training data or need explainability of the model, have a look at:

• Fuzzy Neural Network
• Fuzzification
• Fuzzy Inference
• Defuzzification
• Fuzzy Recurrent Neural Networks

There are multiple neural networks (and their according cells) implemented for time series forecasting:

• Error Correction Neural Network
  • Module
  • Example
  • Reference
• Error Correction Neural Network Cell
  • Module
  • Example
  • Reference
• Error Correction Neural Network ECNN GRU 3 Variant
  • Module
  • Example
  • References
• Historical Consistent Neural Network
  • Module
  • Example
  • Reference
• Historical Consistent Neural Network Cell
  • Module
  • Example
• Historical Consistent Neural Network GRU Variant 3 Cell
  • Module
  • Example
• Deep Historical Consistent Neural Network
  • Module
  • Example
• Causal-Retro-Causal Neural Network
  • Module
  • Example
  • Reference
• Historical Consistent Neural Network With_Know_U
  • Module
  • Example

The Ensemble is applicable for all the above models and a lot more:

• Ensemble

Utils

The utils subpackage contains functions to generate and prepare data for training:

• Create Input for Recurrent Networks
• Create Data for Recurrent Networks

The following utils are interesting for evaluating and investigating your neural network or visualize forecasts with or without uncertainty:

• Neuron Correlation Hidden Layer
• Sensitivity Analysis
• Visualization of Forecasts
• Visualization of Heatmaps