MNIST web

web site for testing neural networks on MNIST data set using 2 hidden layers

To run this you need

node.js
pytorch

Instructions

run node --version on your cmd, if not found go to here and download node.js.
check if you have pytorch installed with this script
```
import torch
print(torch.__version__)
```
run the server by typing node server.js in the cmd.
your server is ready for requests(UI in progress) send a http post request with JSON following the syntax below.

Syntax

General Settings

neural_net - the neural net you want to use out of {"Basic", "Dropout", "Batch_norm", "Combine"}
epochs - number of passes on all the data
learning_rate - learning rate of the network {small values like 0.01, 0.005, 0.001}
batch_size - passing on data using batch_size number of samples each iteration {normally 64}
valid_split - split your train data to validation and training and evalute the network each epoch {range 0 to 1}

Structure Settings

hidden1_size - number of neurons on the first hidden layer {default 100}
hidden2_size - number of neurons on the second hidden layer {default 50}

Options

write_test_pred - write predictions to file 'test.pred', number to represent a boolean(due to a bug in passing a boolean from node.js to python) {0, 1}
draw_loss_graph - graph loss of training and validation, number to represent a boolean(due to a bug in passing a boolean from node.js to python) {0, 1}