Hello!

In this repository I have shared three assignments and two projects of Aipplied Machine Learning course lectured in winter 2018. In the following seections I have given explanation of each project.

1. Linear Regression

This project is composed of 3 sections:

1. Regression: Fitting an n-degree polynomial to the data by minimizing MSE and experimenting effect of L2 regularization.
2. Gradient Descent for Regression: fitting a linear regression model by minimizing MSE with SGD.
3. Experiments on Communities and Crime Data Set from the UCI repository: We had to clean-up the model and handle the missing data by using KNN. Ridge-regression was used to fit a model on the data.

2. Linear Classification and Nearest Neighbor Classification

In this classification project, data we had to generate two classes of data with provided means and covariance using Gaussian distribution. Then we used probabilistic LDA and k nearest neighbored algorithms to reach to the best accuracy and precision. To experiment different shape of boundaries found by probabilistic LDA, we used shared Vs. un-shared covariance for the Gaussian distributions.

3. Sentiment Classification

In this project I implemented a sentiment classifier for Yelp 5 class rating reviews and IMDB binary classification problem. The data was vectorised by first creating a dictionary of words and assigning corresponding vector to each instance of the text corpus. Two vectorization methodologies were experimented including binary representation (existing or not-existing) and also frequency bag-of-words. Linear-SVM, Naive Bayes, and Decision Trees were used to classify these examples.

4. Finding the Largest Digit

In this problem, dataset was randomly generated various grey-scale images containing two or three digits with different sizes, randomly scaled to 40/60/80/100/120 percent of the original digit size. As a data cleaning process, using OpenCV and finding the contours of the digits in the image, the dataset was cleaned from "noise". Different learners were tested, including Neural Networks, Linear SVM, K-Nearest Neighbor, Decision Tree, Random Forests, Bagging, Boosting, Kernelized SVM, and Convolutional Neural Networks (CNNs). For more discussions and information, please check-out our codes and the report.

5. Ablation study of TwinNets

In this project we had to do an ablation study of TwinNet, a bi-directional RNN being tested on sequential MNIST dataset. We also tried all possible machine learning baseline methodoligies as well as hyper-parameter exploration to reproduce the results. For more information, please read the report.