A simple implementation of a feed-forward neural network in D.
This is currently a single layer NN for binary classification using a sigmoid layer and simple delta-based training method, whereby the weights update per entry in the dataset, which are looped over for a number of epochs.
The executable can run directly, or, alternatively, it can be re-built with DUB and run as follows:
git clone [email protected]:patrickm663/feedforward-nn.git
cd feedforward-nn/
dub build
dub run
- Implement logic to split the data into training and testing based on a user's parameter
- Test on more datasets (importing the "wine.csv" dataset to have something to test was one of the biggest hurdles)
- Construct a
neuralnetworkclass that implements the functions as methods rather, in order to create an 'API' - Test whether it can be called from other languages
- Add test cases
- Add additional layers, more activation functions, etc.
MIT licensed.
All contributions are welcome. I am very new to D, but really enjoying the language thus far!