Machine Learning (CS60050): 2023-2024 Autumn

Assignment 1
Submission deadline: 16-Aug-2023, 11:55 PM

* *
Ashwamegh Rathore and Tarun Mohanty
*
Indian Institute of Technology, Kharagpur

Instructions

1. The submission deadline is hard. There may be unforeseen glitches during submission. So,
for safety, submit your files well ahead.

2. All submissions should be on microsoft teams only. No email submission will be accepted.
Special consideration may be made for medical emergencies.

3. Please download the dataset and directory structure to be maintained from here.
dataset
linear-regression.csv
logistic-regression
Pumpkin Seeds Dataset.xlsx
Pumpkin Seeds Dataset Citation Request.txt
4. For each part of the assignment implement the model in a code.ipynb file. Report the model
performance by attaching a report.pdf which contains the output of all cells of the jupyter
notebook. Refer to this article for converting the ipynb file to pdfs. Finally attach the list of
package dependencies for each model in the requirements.txt file.

5. After implementing the required machine learning models, you must submit only the <rollno> <name>
directory. No need to submit the dataset directory. Directory structure to be submitted-
rollno name
linear-regression
code.ipynb
report.pdf
logistic-regression
code.ipynb
report.pdf
requirements.txt
The name of the zip file should be your roll number, followed by an underscore, followed by
your name. For example, if your roll number is 22CS0100 and name is John Doe, then the
zip file should be named as 22CS0100 JohnDoe.zip.
Failing this, your assignment will not be evaluated.

1
 6. Unless explicitly asked,you cannot use any library/module meant for Machine
Learning or Deep Learning. You can use libraries for other purposes, such as formatting
and pre-processing of data, but NOT for the ML part. Also you should not use any code
available on the Web. Submissions found to be plagiarised or having used ML libraries will
be awarded zero marks for all the students concerned.

1 Linear Regression 25 marks

Linear regression is a statistical technique that seeks to establish a linear relationship between a
dependent variable (Y) and one or multiple independent variables (X), by minimizing the sum of
squared differences between actual Y values and the values predicted by a linear equation. It aims
to estimate the coefficients (slope and intercept) of the linear equation, enabling prediction and
interpretation of the dependent variable’s variations based on changes in the independent variables.
For this assignment,

1. Split the dataset into 50% for training, 30% for validation and 20% for testing. Normal-
ize/Regularize data if necessary. Encode categorical variables using appropriate encoding
method if necessary.
(i)
2. Formulate a linear regression model between the observations xj and target parameter
y (i) .
(i)
X
y (i) = θ0 + θj ∗ x j (1)
j

3. Use the mean squared error loss function to fit the 2 following models.

4. For the first model use the analytic solution as discussed in class. Report the R-squared
and RMSE score for the test set after training the model.

5. While for the second model use the an iterative solution via gradient ascent. Use 3 dif-
ferent learning rates- 0.01, 0.001 and 0.0001. Plot the Loss function for the training set and
validation set at each iteration. Finally report the R-squared and RMSE score for test set.

2 Logistic Regression 25 marks

Logistic regression is a statistical method used for binary classification. It models the probability
of an event occurring by fitting a logistic curve to data, assigning values between 0 and 1. It’s a
vital tool for predicting outcomes like yes/no or true/false in various fields.
For this assignment,

1. Split the dataset into 50% for training, 30% for validation and 20% for testing. Normal-
ize/Regularize data if necessary. Encode categorical variables using appropriate encoding
method if necessary.

2. Implement a standard Logistic Regression Classifier as discussed in class. Do NOT

2
 3. Report the mean accuracy, precision and recall for the class 1(good)for the classifiers.
You may or may not use the scikit-learn implementations for computing these metrics.

References

1. Blog on metrics used for bench marking linear regression performance.

2. Blog on metrics used for bench marking classification model performance.

3. Article on exporting package dependencies to requirements.txt

4. Article on setting up a virtual environment using virtualenv

5. Article on generating pdf snapshots from jupyter notebooks.