MNIST web

A website for testing neural networks on MNIST data set using 2 hidden layers.

Required Software

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__)
```
Install required packages(while in the directory)
```
npm install
```
Run the server by typing node server.js in the cmd.
Your server is now ready for requests.
In the browser type http://localhost:8000/, fill the form and send.
Wait for network's results.

Syntax

General Settings

neural_net - The neural net you want to use out of {"Basic", "Dropout", "Batch_norm", "Combine"}.
epochs - The number of data iterations.
learning_rate - The network's learning rate {small values like 0.01, 0.005, 0.001}.
batch_size - Iterating the data using batch_size number of samples each iteration {normally 64}
valid_split - Splits your train data to validation and training and evalutes the network after each epoch {ranges from 0 to 1}.

Structure Settings

hidden1_size - The number of neurons in the first hidden layer {default is 100}.
hidden2_size - The number of neurons in the second hidden layer {default is 50}.

Options

write_test_pred - Writes 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 - Draws loss graph of training and validation, number to represent a boolean(due to a bug in passing a boolean from node.js to python) {0, 1}.