README.md

README
Tested on Ubuntu 20 and python>=3.8
This work is done for the CSE600A (Object Oriented Programming and Design) course requirement for group number 41. The problem statement is given in problem.pdf.

Overview

1. In this work we implement a simple 2 layer feed-forward arrtificial neural network (ANN) from scratch using numpy only. Gradients are calculated are calculated similarly to this work: http://cs231n.stanford.edu/slides/2021/discussion_2_backprop.pdf.
2. Doxygen file is at latex/refman.pdf
3. For the profiler file refer to the test/test.txt
4. ann/example2l (2l = 2 layer) class has the implementation of training and testing of the 2 layer network.
5. We achieve an accuracy of 95 percent when trained for 1 epoch.
6. Data and output predictions are in data folder (after you erxtract the data.tar.gz).

Model specifications

i_d = 784 # input dimension    
m_d = 100 # no of cells in the mid layer    
o_d = 10 # output dimension    

epoch = 1 # we train for 1 epoch only    
act = 'relu' # activation type      
lr = 0.00001 # learning rate    

X,Y,X_test,Y_test = train_features, train_label, test_features, test_label  

Run these commands to set up the environment. Conda need to be installed. If not than you have to manually install the dependencies.

conda env create -f environment.yml    
conda activate oopd     
python setup.py sdist bdist_wheel      
pip install dist/ann-0.0.1-py3-none-any.whl    

Run these commands to extract the MNIST dataset and to train the example network on it.

tar -xvf data.tar.gz    

This will extract the database required to train the network on MNIST dataset.

python run.py  

After the traning the predicions and their true values will be saved in data/test_pred_true.csv file. To conver this into a sql database run the below commands.

Run these commands to save the prediction to a sql database.

.mode csv 
.import data/test_pred_true.csv oopd  
.save data/oopd_pred_true.db