Programming Assignment

getnaveens

CS6375: Machine Learning (Spring ‘23)
Programming Assignment #1
Logistic Regression
DUE: Friday February 17th at 11:59 PM
Problem: Implement the Logistic Regression algorithm. The code skeleton as well as data sets for this
assignment can be found on e-Learning.
Data Set: The data set (in the folder ./data/) is obtained from the UCI Repository and are collectively the
MONK’s Problem. These problems were the basis of a first international comparison of learning algorithms1. The training and test files for are named monks-3.train and monks-3.test. There are six
attributes/features (columns 2–7 in the raw files), and the class labels (column 1). There are 2 classes.
Refer to the file ./data/monks.names for more details.
a. (Reporting Error Rates, 30 points) For the following learning rates {0.01, 0.1, 0.33} and
iterations {10, 100, 1000, 10000} fit Logistic Regression Models using the MONKS-3 train set and
compute the average training and test errors. Report what the best parameters are for the model.
b. (Saving the Model, 50 points) Retrain the models on the best parameters and save it as a pickle
file in the format ‘NETID_lr.obj’. The object file will be loaded for grading and the class functions
will be individually tested.
c. (scikit-learn, 10 points) For monks datasets, use scikit-learns’s default Logistic Regression
Algorithm2. Compute the train and test errors using sklearn’s Logistic Regression Algorithm.
Compare the results with the version you implemented and speculate on why there is a difference in
performances between the two algorithms. Do not change the default parameters.
d. (Plotting curves, 10 points) For each of the learning rates, fit a Logistic Regression model that
runs for 1000 iterations. Store the training and testing loss at every 100th iteration and create three
plots with epoch number on the x-axis and loss on the y-axis.
.
1https://archive.ics.uci.edu/ml/datasets/MONK’s+Problems
2 https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html
Programming Assignment # 1