Statiscal Methods in Aritifical Intelligence

Machine Learning And Deep Learning Algorithms implemented from scratch

All algorithms mentioned below are implemented from scratch.

Assignment 1 contents:

1. KNN Classifier

    Implemented a KNN Classifier from scratch for all values of K and different distance measures.
    Trained and tested the classifier using MNIST dataset and reported all the metrics.
    Compared the classifier with the scikit-learn's built-in KNN classifier.
    
   Used the same classifier on the Mushroom dataset.
   Used plots to show how the missing data is handled.

2. Decision Tree Classifier

    Implemented Decision Tree Classifier from scratch to predict housing prices (attributes - numeric+categorical).
    Normalized data and used variance reduction as the criterion for choosing split in the decision tree.
    Reported metrics such as MSE and MAE and compared the classifier with scikit-learn's built-in Decision Tree Classifier

Assignment 2 contents:

1. Linear SVM

    Implemented a Linear SVM Classifier from scratch for prediction of test images on CiFAR dataset.
    Ipynb file contains a brief description of SVM and its parameter (eg. significance of C)
    Compared the classifier with the scikit-learn's built-in KNN classifier and reported the support vectors.

2. Guassian Mixture Models Clustering

3. Linear Regression Model

    Implemented Linear Regression Model from scratch to predict apparent temperature in the given dataset of historical weather..
    Reported metrics such as MSE and MAE and reported the behavior of the model on different regression coefficients. 
    Used different plots to show the change in MSE and MAE for different number of iterations in conditions like:
            a. Different coefficient values for a fixed learning parameter.
            b. Different learning parammter values for a fixed coefficient value.

4. KMeans Clustering

    Implemented KMeans Clustering from scratch for clustering the given dataset of documents from 5 different fields
    Used pretrained model and vectorizers (eg TF-IDF) for feature extraction from the documents

Assignment 3 contents:

1. PCA

    Performed PCA on the image dataset. Downscaled the image to grayscale for better results and ease of computation.
    Reconstructed the image with small number of components.
    Showcased the variation of MSE of the dataset with the number of components using plots.
    Used scatterplots to show the varaition of clusters in 1D, 2D, 3D planes with the required number of components obtained from PCA.

2. Logistic Regression Model

   Implemented the logistic regression model.
   Using numpy, extracted the eigen values and vectors and reported accuracy metrics.

3. Multilayer Perceptron

   Implemented MLP, CNN and SVM to classify images into their respective category.
   Plotted graphs for different scenarios and summarized them:
            a. Number of hidden layers
            b. Different activation functions
            c. Different kernels (SVM) 

Assignment 4 contents:

Used the algorithms implemented above in real world datasets

    Participated in different contests on Ai-Crowd using the above implemented algorithms
    Questions: 1. Hate Speech Detection Challenge
               2. Tom and Jerry Emotion Detection Challenge
               3. Covid Drug Discovery for Covid19