Recurrent Neural Net Versions Save

A recurrent (LSTM) neural network in C


3 years ago

This version contains some corrections and improvements from the previous release:

  • storing .net files during training (-st flag) now also includes the header section
  • added new flags -vr and -c
  • added github integrated CI to the repository (workflows, travis)
  • Added an example in the that illustrates how one can train the same network on multiple files.

Happy Holidays Rickard


3 years ago

From here on, binary files have a supplied header section. It makes previous binary files invalid, but will handle things much better in the future.

Major change is that the feature set is stored in the binary file, making it possible to train the same network over more than one datafile. The problem of having all data in memory is thus overcome, finally.

In the back seat of the car, using my phone as a personal hotspot, on my way to celebrate Jul. Merry Christmas! Rickard


4 years ago

For future reference, this version will be tagged as v1.0. I have built it for Windows, Linux and macOS.

It works as a command terminal program used for getting the weights of the network. These weights are stored in either ascii format (JSON) or as raw bytes. The HTML application can be used to as inspiration on how these weights can be put into use.


Consult the README file.