Neugrad

This is a small autograd engine, made purely from numpy and python. This is still a work in progress and more features will be added with time Hope you have fun implementing this :) Additionally, later on, deep learning layer features, loss functions and optimizers will be added to this project. This aims to be a small pytorch clone

Requirements to run it

It would help if you had these installed in your system before running this engine

• Numpy
• Typing
• Python (version 3.7+)

How to run it

• Download all the files (only engine.py)
• Use this syntax from <file location of this folder> import * to import it into your code
• You can also see the tests I run on the engine in test.py

Features added till now (with backpropagation)

Scalar and Tensor Operations

• Creation of Tensor (NVal)
• Addition
• Subtraction
• Negative of Tensor
• Tensor Power (Aᴮ) ~ B is a scalar
• Scalar Multiplication
• Scalar Division
• Max Function
• MatMul Function
• Transpose
• Mean
• Greater Than
• Lesser Than

Activation Functions

• ReLU
• Sigmoid
• Tanh

Helper Functions

• Randn and Randn Like
• Rand
• Randint and Randint Like
• zeros and zeros like
• ones and ones like
• max and argmax
• min and argmin

Converting Functions

• to array
• to NVal
• to list

Deep Learning Operations

• Linear Regression
• Dropout
• Conv2d
• Max2d

Optimizers

• Adam
• SGD

Loss Functions

• BCE Loss
• MSE Loss
• Hinge Loss

Active Bugs

• None

Please keep in mind this is a work in progress and will keep on being updated over time.