Skip to content

Repository files navigation

Overview

A simple implementation of a sequential neural network API using Numpy.

Implemented following objects: Dense layer, MSE loss, ADAM optimizer.

Breakdown

Optimize1-4 are iterations towards Optimize5.

Optimize5 contains the main classes.

Optimize6 contains the testing/demo program.

Demo.mov is a run of Optimize6.py

Additional notes(pdf and png) are provided towards the developement of this program

Extra Info

Multiple layers can be chained to a model using Layer().add_to(model)

The dense layer was efficiently implemented by exploiting the underlying symmetry of derivatives.

A sample model was generated and used to approximate a linear combination, average, and polynomial.

About

My implementation of a custom sequential neural network using gradient. I used the natural symmetry of weights to efficiently store the values. In this project, I have implemented the following: Input-Dense-Output layers, MSE loss, and the ADAM optimizer.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages