OVS_lab1: Neural Network Implementation

This repository contains a C++ implementation of a neural network designed for pattern recognition tasks. The project is structured to facilitate both training and inference operations, with support for custom datasets.

• Custom Neural Network: Implemented from scratch in C++.
• Training and Inference: Capable of training on labeled datasets and performing inference on new data.
• Dataset Generation: Includes scripts for generating training and testing datasets in a specific format.
• Evaluation Metrics: Outputs training time, number of epochs, inference time, and memory usage during training and inference.

OVS_lab1/
├── CMakeLists.txt # Build configuration
├── main.cpp # Main application entry point
├── include/ # Header files
│ └── neural_net.h # Neural network class and utilities
├── src/ # Source files
│ └── neural_network.cpp # Implementation of neural network
├── data/ # Dataset files
│ ├── train/ # Training samples
│ ├── test/ # Test samples
│ └── model # Neural network model
└── README.md # Project documentation

• C++17 or later
• CMake 3.10 or later
• Python 3.x (for dataset generation)

1. Clone the repository:

   git clone https://github.com/Vsevomes/OVS_lab1.git
   cd OVS_lab1

2. Create a build directory and compile:

   mkdir build
   cd build
   cmake ..
   make

3. The build will produce an executable file (e.g., neural_network) in the build directory.

You can use the executable to either train a model or predict using a saved model.

1. Train a Model

   ./neural_net train <train_folder> <num_inputs> <hiddenLayers> <neuronsPerLayer> <num_outputs> <learningRate>

   • <train_folder> – folder containing training files
   • <num_inputs> – number of input neurons (e.g., 49 for 7×7 images)
   • – number of hidden layers
   • – number of neurons per hidden layer
   • <num_outputs> – number of output neurons/classes
   • – learning rate for training

   Example:

   ./neural_net train ../data/train/ 49 1 20 3

2. Predict Using a Saved Model

   ./neural_netw predict <model_file> <input_folder>

   • <model_file> – path to the saved model file
   • <input_folder> – folder containing test files

   Example:

   ./neural_net predict model.dat ../data/test/

Each dataset file should contain:

1. The first line: a space-separated vector representing the one-hot encoded target class.
2. Subsequent lines: a 7x7 grid of binary values representing the input pattern.

Example:

0 0 1
0 0 0 0 0 0 0
0 0 0 1 0 0 0
0 0 1 0 1 0 0
0 1 0 0 0 1 0
1 0 0 0 0 0 1
1 1 1 1 1 1 1
0 0 0 0 0 0 0

Python scripts are provided to generate training and testing datasets:

python generate_datasets.py

This will create train/ and test/ directories populated with sample files.

Upon execution, the program will output:

• Training Time: Duration of the training process.
• Epoch Count: Number of epochs completed.
• Inference Time: Time taken to process the test dataset.
• Memory Usage: Estimated memory consumption during training and inference.