This repo consists of codes and reports made as part of the graduate level course at McGill University. Train and test files of datasets (csv and/or pkl), final Python notebooks, and PDF reports in the format of a NeurIPS 2020 LaTeX Template for each mini project have been added to respective folders.
We implemented the linear classifier - Logistic Regression - from scratch in Python. The model was trained on two given datasets and goal was to perform accurare binary classification on them both.
We investigated the performance of various classifiers on a mutli-class text classification problem. We developed the Bernoulli Naive-Baiyes Algorithm from scratch, and compared its performance with five classifiers from SciKit-Learn package: Logistic Regression, Neural Network, Decision Trees, Random Forest, and Support Vector Machines (SVM). Also participated in an internal Kaggle competition and achieved 0.94611 test accuracy in final competition.
We investigated the performance of two deep convolutional networks: LeNet and ResNet on a 10-class image classification problem. By training the models and evaluating their performance for varying epochs, employing data augmentation, and dropout regularization, we observed that ResNet outperfomed LeNet. Furthermore, ResNet-34 exhibited better performance than ResNet-18. Also participated in an internal Kaggle competition and achieved 0.97614 test accuracy in final competition.