Gradient Descent

Description

This project explores how we can use supervised learning to solve a regression problem; predict the sale price of a house given a particular age. The algorithm chosen here is gradient descent with univariant linear regression using the mean squared error as the cost function.

Common Terms

Gradient Descent is a machine learning algorithm which is used to find the local minima (or global minima in convex functions) of a given cost function.

Machine learning is the science of getting computers to learn without being explicitly programmed. - Arthur Samuel (1959)

Supervised learning is when you give the computer all the input, output pairs for the training data.

Regression problem - Predicting continuous values

Getting Started

To install dependencies we use yarn:

yarn install

Starting locally:

To start the server:

yarn workspace server dev

To start the app:

yarn workspace app dev

Lerna

First off, What is lerna? lerna is a tool that allows you to maintain multiple npm packages within one repository.

There's a couple of benefits to this kind of approach, the paradigm is called a monorepo, and more can be read about it from the source of babel, and react.

Here's the gist:

• Single lint, build, test and release process.
• Easy to coordinate changes across modules.
• Single place to report issues.
• Easier to setup a development environment.
• Tests across modules are ran together which finds bugs that touch multiple modules easier.

Vite

Vite is a new and upcoming frontend build tool which is blazingly fast and comes with rich out of the box features such as a ready to go static file server. For more information about vite visit: https://vitejs.dev/.

TODO:

• Convert csv data to json
• Setup a simple server for client to retrieve data
• Plot the data onto a scatter diagram
• Plot the hypothesis line
• Calculate cost function
• Gradient descent algorithm
• Contour plot of the error
• Enable controls for adjusting the parameters of the line
• Visualise the steps of GD alg