📈 : Linear-Regression
A simple program where you can plot lines and see a linear regression algorithm find the line of best fit!
Usage
To run, simply download the simulator.zip, extract the file, open it and then click on the simulator application.
- Download and extract Zip
- Run simulator application
Main features
- clickable graph to input points that can be cleared at any time
- simple linear regression framework using pytorch
- Linear regression metrics
- Epochs - Number of training runs the model has had
- Training loss - The loss between estimated and real training values in raw pixelss calculated using Huber Loss
- Test loss - The loss between estimated and real test values in raw pixels
- R squared - A metric that ranks how close the model is to a line that perfectly crosses through all dots (from -1 to 1)
- Equation of line in pixel values
Running in python
If you are running the source code, you need to ensure that you have python 3, pygame and pytorch installed, and run the main simulator.py
How it works
The Linear regressor is a simple torch network with 2 parameters, weight (w) and bias (b), and calculates a value by plugging the input (x) into the formula: $ wx + b$
The training data and test data is randomly chosen from the nodes you click on the graph and every 10 epochs the model displays it training progress.
Credits and Acknowledgements
AI declaration: I used gemini for bugfixing and feedback, and it taught me a lot about normalisation
Credit to learnpytorch.io which I used to learn how to make a linear regressor in pytorch
Credit to pytorch and pygame libraries that I used for my project