Scribd
Upload
47 views

Supervised Assignment 2

The document provides instructions for an assignment to build a neural network in Python from scratch to classify 2D points into two classes based on their x and y coordinates, with the neural network required to have at least one hidden layer with a minimum of 4 neurons and be trained over 100 epochs with a learning rate of 0.005 using mean squared error as the error function and sigmoid activation for the output layer. No libraries can be used except for math, time and random for specific functions.

Uploaded by

Ahmed Kadry
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
47 views

Supervised Assignment 2

The document provides instructions for an assignment to build a neural network in Python from scratch to classify 2D points into two classes based on their x and y coordinates, with the neural network required to have at least one hidden layer with a minimum of 4 neurons and be trained over 100 epochs with a learning rate of 0.005 using mean squared error as the error function and sigmoid activation for the output layer. No libraries can be used except for math, time and random for specific functions.

Uploaded by

Ahmed Kadry
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Cairo University Faculty of Computers and Artificial Intelligence

Third year (2020-2021) Supervised learning (Spring 2021)

Assignment 2
Programming Language: Python

Input: Draw some points on a 2D grid for two classes, at least 6 points for each class, most of
those points should be distinguishable as class 1 or class 2 (choose the points yourself). The
coordinates (x, y) will be your input (x1, x2), the class that the point belongs to will be either 0
or 1. The output class is what your neural network should learn to classify. (NOTE: the drawing
part is for yourself, you can do it on a paper or paint to determine where your points will be
and to make sure that most of them can be distinguished)

Requirement: Build your own neural network using only python’s available built-in functions, without
any libraries/imports (3 helper libraries written in the restriction part). The neural network should
contain at least 2 layers: 1 hidden, 1 output.

An example: input hidden layer 1 hidden layer 1 output  output layer  output.

The number of neurons in a layer should be of a minimum 4. Do note that the number of neurons is NOT
fixed, do not write your code based on a fixed number, it should be dynamic and changeable.

Features: Your input is as explained above, and the output is 0 or 1 depending on the class. Note that
you threshold the output to classify, but you don’t do that in back propagation as you need the error
value.

Prediction: The neural network will do the predictions for you. You just have to threshold the value
after using the activation function to get the prediction. Compare the prediction with the original and
get %.

Summarized Steps: You will use the equations that you have taken in the lecture to pass the input
through your neural network layers. You will initialize the weights of the layers with random values. You
have the equations to calculate the output for all the forward passes which is simple multiplication of
the values and the input values/values from previous layer. Do note to save the output values as you will
need them for the back propagation.

Your activation function for the final output layer will be the sigmoid function, and your error function
will be MSE. For the back propagation you can also refer to the lecture to know how to implement it.
You will propagate back the error and change the weights accordingly. Your learning rate should be set
to 0.005, run the learning for 100 epoch. 1 epoch means a forward and backward propagation of all your
samples.
Restrictions: No imports should be used except for math, time and random library are the exception
and those are included for the use of ceil, floor, round, time and random. You should build the neural
network and calculate everything that happens in it yourself using lists, arrays, for loops, etc… no
shortcuts allowed.

You might also like