Skip to content

Latest commit

 

History

History
101 lines (81 loc) · 3.8 KB

README.md

File metadata and controls

101 lines (81 loc) · 3.8 KB

Welcome to the Repository!

This repository contains 8 projects, each focusing on a unique machine learning task. These projects explore data analysis, regression, classification, dimensionality reduction, clustering, and deep learning using diverse datasets. Below is a detailed description of each project and its objectives.

Project 1&2: Prediction

Objective:

Build a regression model to predict attributes such as GDP per capita, social support, freedom, and other socioeconomic indicators.

Key Tasks:

  1. Data Analysis:
    • Summarize the dataset (size, feature types).
    • Visualize distributions and analyze correlations.
  2. Feature Engineering:
    • Compute statistical values and identify features requiring special treatment.
  3. Model Training:
    • Train and evaluate linear regression models using closed-form solutions and SGD.
    • Explore Ridge, Lasso, and Elastic Net regularization.
    • Compare performance with polynomial regression.
  4. Evaluation:
    • Test the model using a separate test set and evaluate its performance.

Tools:

  • Jupyter Notebook for analysis and visualization.
  • Libraries: NumPy, Pandas, Matplotlib, Scikit-learn.

Project 3&4: Rock Classification

Objective:

Classify rock types (Igneous, Metamorphic, Sedimentary) using a dataset with 11 features.

Key Tasks:

  1. Data Preparation:
    • Process and split the dataset into training, validation, and testing sets.
    • Visualize feature distributions and analyze correlations.
  2. Model Training:
    • Train classifiers such as Logistic Regression, SVM, and Random Forest.
    • Tune hyperparameters using grid search or manual exploration.
  3. Ensemble Methods:
    • Combine individual classifiers into an ensemble for improved performance.
  4. Comparison with Human Performance:
    • Compare model accuracy with human accuracy and analyze the results.

Tools:

  • Jupyter Notebook for analysis and training.
  • Libraries: Scikit-learn, Matplotlib, Seaborn.

Project 5&6: PCA and Clustering on Rock Images

Objective:

Apply PCA and other dimensionality reduction techniques to analyze rock images and identify clusters.

Key Tasks:

  1. PCA Analysis:
    • Determine the number of components required to retain 90% variance.
    • Reconstruct and visualize images with reduced dimensionality.
  2. Visualization:
    • Use PCA, t-SNE, LLE, and MDS to create 2D scatter plots of image embeddings.
  3. Comparison with Human Features:
    • Use Procrustes analysis to compare embeddings with human-labeled features.
  4. Clustering:
    • Perform clustering using K-Means and EM.
    • Visualize and evaluate clustering performance.
  5. Neural Networks:
    • Build a feedforward neural network (using dense and CNN layers).
    • Procrustes analysis with human data.

Tools:

  • Image processing: OpenCV, Matplotlib.
  • Dimensionality Reduction: Scikit-learn.
  • Procrustes Analysis: SciPy.

Project 7&8: Working with Large Language Models

Objective:

Evaluate the performance of a pre-trained models on dataset.

Key Tasks:

  1. Model Description:
    • Analyze the architecture and parameter breakdown of the selected CLIP model.
  2. Image Classification:
    • Evaluate model accuracy across five conditions (realistic, geons, silhouettes, blurred, features).
    • Compare results with human performance.
  3. Dimensionality Reduction and Visualization:
    • Use t-SNE to visualize embeddings in 2D space.
    • Add object images to the embedding visualization.
  4. Analysis:
    • Compare model embeddings with human-labeled features using Procrustes analysis.
    • Compute correlation coefficients and analyze significance.

Tools:

  • Pre-trained models: Huggingface Transformers.
  • Visualization: Matplotlib, t-SNE.
  • Libraries: NumPy, SciPy, Scikit-learn.