This folder will [hopefully] contain some implementations of different Neural Networks.
Currently it only contains basic implementation of backpropagation algorithm for 3 layers Neural Network classifier for 10 different labels. Current implementation does not allow for differnt NN architectures but it might do in the future.
Backpropagation NN example uses a sample of 5000 images from MNIST database for training and validation. It performs classification of digits from 0-9.
You must have R installed on your computer. Provided scripts were tested on
R version 3.2.2 (2015-08-14) -- "Fire Safety".
There is a simple R script (
NNClassify.R) provided to demonstrate the NN training and validation. The script accepts following cli parameters:
- path to a training set CSV file
- type of the training set
- normalize - TRUE/FALSE - do you need to scale the features in the training set?
- lambda - regularization parameter. If not specified regularization is not performed
Rscript nnClassify.R "data/data.csv" "csv" FALSE 1 Loading training set: data/data.csv Done! Computing BFGS optimized NN parameters with 50 iterations initial value 7.038124 iter 10 value 1.814019 iter 20 value 1.099056 iter 30 value 0.814639 iter 40 value 0.669605 iter 50 value 0.580098 final value 0.580098 stopped after 50 iterations Done computing NN parameter! Neural Network classification accuracy: 93.4