Descrição
this is Layer Base NeuralNetwork Blueprint Plugin.
you can build a NeuralNetwork using several kinds of Layers.
Main Class of this Plugin is "NglSimpleLayerStackNN".
this class have some methods for "Build Network", "Train" and "Inference".
"Layer Class" is Part of NglSimpleLayerStackNN. There are 8 different Layer Classes.
NglSimpleLayerStackNN is constructed by adding several Layers.
NglSimpleLayerStackNN processes the Stacked Layer in order from the input side to the output side.
you can Train Game AI as follows.
https://youtu.be/JMm5Z8aykXc
The explanation is the following URL.
https://github.com/nagakagachi/ue4/wiki/NglNN-Sample-2
Note: GPU acceleration is not supported.
Layer Classes
- Affine
- Convolutional
- GRU (Gated Recurrent Unit)
- Softmax
- Sigmoid
- ReLU
- Softplus
- Tanh
NglSimpleLayerStackNN
- Main Class of this Plugin.
- "Construct Network" : create "Layer Class" and add to Network by "AddLayer" method.
- "Train" : Call "Train" method with Input-Data and Teach-Data.
- "Inference" : Call "Forward" method with Input-Data to Calculate Output.
- "Save" : "SaveJson" method write network parameter and structure to Json.
- "Load" : "LoadJson" method construct network structure and read parameter from Json.
Documents
https://github.com/nagakagachi/ue4/wiki/NglNN
https://github.com/nagakagachi/ue4/wiki/NglNN-Sample-1
https://github.com/nagakagachi/ue4/wiki/NglNN-Sample-2