CMPE 462 - Assignment 1 Solved

Introduction
This assignment consists of 2 parts.

The first one is about percentrons and the second part is about linear regression.

Part1
In Part 1 of this assignment, you will implement Perceptron Learning Algorithm (PLA) from scratch (details on page 15 of Lec02-Perceptron notes) and work on 2D data.

Given the target separating function f as below:

y = -3x+1
you will generate 2D data points (x,y) in two classes {0,1} defined as:

if y < -3x+1, then c=0 if y > -3x+1, then c=1
You will run your PLA in order to classify the generated points and plot the resulting decision boundary.

The steps of Part1 are below:

•   Step1: Generate total 50 points in two classes and apply PLA.

•   Step2: Generate total 100 points in two classes and apply PLA.

•   Step3: Generate total 5000 points in two classes and apply PLA.

For each step, plot the target separating function f (in green) and the generated points (use red for class 0, blue for class 1) and your decision boundary (in purple). Use labels for both axis and the target separating function f and your decision boundary. Save your plot as a png file named as part1_stepn.png, example part1_step1.png.

You can use mathplot library. But special functions or libraries for decision boundary plotting is forbidden.

In your project report, include results of your runs for each step, 50,100 and 5000. Place the plots and discuss over them. Compare your resulting boundaries and the target function. Also discuss the number of iterations you need for each step.

Part2
In Part 2 of this assignment, you will implement Multiple Linear Regression from scratch (closed form solution is on page 41 of Lec03-LinearRegression notes). In Linear Regression, the relation between an input variable (1D) and an outcome (1D) is modelled. In Multiple Linear Regression, the input variable is multidimensional and the modelled relation is as below:

y=c1∗ x1+ ... + cn ∗ xn + e
where y is the outcome (dependent variable), xi are the input parameters (independent variables), ci are the coefficients and e is the error term.

For Part2, two input datasets (DS1 and DS2) will be given to you. Each dataset file is a csv file with no header, no index. Each row represents a different sample. Each column in a sample represents a variable. The last column keeps the dependent variable. Below is an example csv file. There are 3 lines, it means that there are 3 samples. There are 6 columns in each line, it means there are 5 independent variables (first 5 columns in a row) and 1 dependent variable (last column). In this example all dependent variables are 6.

1, 2, 3, 4, 5, 6

1, 2, 3, 4, 5, 6

1, 2, 3, 4, 5, 6
The steps of Part2 are below:

•   Step1: Run Multiple Linear Regression on DS1.

•   Step2: Run Multiple Linear Regression on DS2.

•   Step3: Implement Multiple Linear Regression with l2 regularization (details on page 50 of Lec03-LinearRegression notes). Run Multiple Linear Regression with l2 on DS2.

For each step, plot the loss over iterations. Save your plot as a png file named as part2_stepn.png, example part2_step1.png. If you implement closed form solution, you can print the time taken for your run in miliseconds as: "Time to complete stepn: XX msec", example "Time to complete step1: 550 msec"

In your project report, include results of your runs for each step. Place loss over iterations plots and discuss over them. Also discuss the number of iterations you need for each step. If you implement closed form solution, you can discuss the durations for your runs.

Base Environment
You will be implementing your code with Python 3.6.

You need to create a python virtual environment with Anaconda for your project. After installing Anaconda, a base environment can be created with below commands:

conda create -n 462assignment python=3.6 conda activate 462assignment
While you keep working on your models, you will need to import additional libraries. List these libraries in a requirements.txt file. State any special versions if needed. A sample requirements file can be as below:

scikit-learn >= 0.22.2 scipy pandas sentencepiece==0.1.91
For grading, we will load your requirements with the command below:

python3 -m pip install -r requirements.txt
Before submission, test your code on a clear new conda environment by installing additional libraries from your requirements file. Because, there will be penalty if your code doesn’t run like this.

Grading Details
The assignment will be graded over 100 points. You will be graded for your code and report.

•   10 points for report

•   90 points for code

–    45 points for Part1

∗ 15 points for step 1 ∗ 15 points for step 2

∗ 15 points for step 3

–    45 points for Part2

∗ 15 points for step 1

∗ 15 points for step 2

∗ 15 points for step 3

We will run your code on a clear new conda environment. First we will load your requirements.txt file. Then we will test your code with below commands:

•   Part1

python3 assignment1.py part1 step1 python3 assignment1.py part1 step2 python3 assignment1.py part1 step3
Consider first command, you will generate 50 2D data points, run PLA, generate your plot and save it as png file.

•   Part2

python3 assignment1.py part2 step1 python3 assignment1.py part2 step2 python3 assignment1.py part2 step3
Consider first command, you will run linear regression on DS1, generate your plot and save it as png file.