satyaki0924 / minilearn

Mini-learn is a miniature version of tensor-flow which I made with ONLY NUMPY to play with perceptrons. You can use this project like you use tflearn. Go to https://github.com/Satyaki0924/boston-housing-with-minilearn to see it's usage.

Minilearn

A Miniature tensorflow made for experimentation

Classes:

Node:

Parent class. The constructor is used to set properties for all the subclasses which inherits from this class.

Input:

The Input class with which we will input the data to mini-learn

Linear:

performs linear transformation using the formula Y = mX + b

Sigmoid:

Represents a node that performs the sigmoid activation function. We use the formula σ = 1 / 1 + e**-x

Mean Squared Error (MSE):

The mean squared error cost function. More about MSE: https://en.wikipedia.org/wiki/Mean_squared_error

Functions:

A class consisting of several static functions:

topological_sort:

Sort the nodes in topological order using Kahn's Algorithm.

forward_and_backward_pass:

Performs a forward pass and a backward pass through a list of sorted Nodes.

sgd_update (Stochastic gradient descent):

Updates the value of each trainable with Stochastic gradient descent

Author: Satyaki Sanyal

This project must strictly be used for educational purposes only