NeuralNetworkFromScratch

I decided to create a simple back propagation neural network from scratch in C++ as part of my final project for my Scientific Computing course (DATASCI 2G03).

Instructions for Use

Read the project proposal to learn about the model and how it works!

1. Unzip the training data and training labels
2. Enter the paths to each respective dataset file in demo.cpp
3. Run make
4. Run ./demo and the output to the terminal will describe the program as it runs.

Environment Information

• Running on CentOS Linux 7 (Core)
• gcc (GCC) version 4.8.5 (Red Hat 4.8.5-44)

Sample Output

Opening path-to/t10k-images.idx3-ubyte...
Magic number: 2051
Number of images: 10000
Number of rows: 28
Updated image number of rows to: 28
Number of columns: 28
Updated image number of columns to: 28
Initializing Inputs Vector...
Loading images...
Success!
Updated dataset size to: 10000
Opening path-to/t10k-labels.idx1-ubyte...
Magic number: 2049
Number of labels: 10000
Loading labels...
Success!
Updated dataset size to: 10000
Opening path-to/train-images.idx3-ubyte...
Magic number: 2051
Number of images: 60000
Number of rows: 28
Updated image number of rows to: 28
Number of columns: 28
Updated image number of columns to: 28
Initializing Inputs Vector...
Loading images...
Success!
Updated dataset size to: 60000
Opening path-to/project/train-labels.idx1-ubyte...
Magic number: 2049
Number of labels: 60000
Loading labels...
Success!
Updated dataset size to: 60000
Initializing network with input layer size 784, hidden layer size 10, and output layer size 10...
Beginning epoch 0
Running training batch 0 to 9999
Initializing vectors for storing layer outputs...
Forwards propigation...
Current accuracy on training data...
Accuracy: 0.0507
Current accuracy on validation data...
Accuracy: 0.0539
Backwards propigation...
Applying changes...
...(continued for 10 epochs)...
Running training batch 50000 to 59999
Initializing vectors for storing layer outputs...
Forwards propigation...
Current accuracy on training data...
Accuracy: 0.7744
Current accuracy on validation data...
Accuracy: 0.7606
Backwards propigation...
Applying changes...