Repository files navigation

Erlang-Shen is a naive concurrent implementation of the neural network learning algorithm in Erlang byAlex Lenail andSunjay Bhatia for COMP 50-02 Concurrent Programming at Tufts University. It is named after the Chinese truth-seeing god,Erlang Shen. Through this implementation, we hope to use the power of concurrency in Erlang to take a different approach to neural networks and model them in a way that is more analagous to how the brain functions. Our network currently can only handle binary class predictions, but we hope to expand on this in the future to handle arbitrary ARFF data and support a distributed neural network.

    git clone https://github.com/zfrenchee/Erlang-Shen.git    cd Erlang-Shen

    make clean && make

    erl -pa ebin

    application:start(shen).    shen:run(TrainingDataFile, TestDataFile, HiddenLayerDimensions, GradientDescentSteps).

WhereTrainingDataFile andTestDataFile are strings representing paths to valid ARFF format data files. Example files can be found in thedatasets folder of this repository.HiddenLayerDimensions is a list of integers that can be specified as the hidden layer architecture of the neural network.GradientDescentSteps is an integer specifiying the number of gradient descent steps to take to tune the network. Results are displayed and output to theresults folder in this repository. For best results on the Iris dataset, it is recommended that you use a single hidden layer of size 4 (the number of features in the dataset) and at least 400 gradient descent steps. We have found with these parameters, we can get 94% accuracy on the Iris dataset.

    application:stop(shen).