Neural Network Approximation of sin(x)

Description

This repository contains Python code for a neural network that approximates the sin(x) function.

The code demonstrates how to:

• Load and prepare PyTorch tensors for data.
• Split data into training and testing sets.
• Create a data loader for batch processing.
• Define and train a neural network model.
• Evaluate the model's performance on test data.
• Make predictions using the trained model.
• Plot the actual sin(x) function and the model's predictions for visualization.

Model summary (generated by torchsummary)

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Linear-1                  [-1, 500]           1,000
            Linear-2                 [-1, 1000]         501,000
            Linear-3                    [-1, 1]           1,001
================================================================
Total params: 503,001
Trainable params: 503,001
Non-trainable params: 0
----------------------------------------------------------------

Requirements

• Python 3.x
• PyTorch
• scikit-learn
• Matplotlib
• torchsummary (optional)

For exact versions, check requirements.txt.

Instructions

git clone https://github.com/g-nitin/nn-sin-approximation.git
cd nn-sin-approximation
pip install -r requirements.txt

python3 main.py

Directory Structure

• sin_network.py contains the network to approximate sin and functions to train and evaluate it.
• utilities.py contains helper functions to plot the predictions.
• main.py contains the main function.
• The plots subfolder contains different plots created by various training sizes.
• The models subfolder contains the various models based on the training size.

Results

Actual v. Predicted sin(x)