Machine Learning Case Studies

Description

Document Retrieval using Clustering and Similarity:

• Used the k nearest neighbors algorithm for document retrieval of similar documents.
• Calculated similarity using tf-idf for every document present in the corpus and the distance between two documents using cosine distance.
• Performed queries such as top 5 documents closest to the given document.

Image Retrieval using Transfer and Deep Learning:

• Used transfer learning technique to extract the deep features of the image-net model and trained a k nearest neighbors model with these extracted deep features.
• Performed image retrieval by calculating the distance between any two images using the trained model.

Personalized Song Recommender using Item Similarity:

• Created a simple popularity based recommender based on total times each song has been played and personalized song recommender and concluded the latter to be more accurate.
• Performed analysis to find most recommended song, similar songs, and recommended each user with a different list of songs based on the past history of songs the user has listened to.

Code

1. Clustering & Similarity - Document Retrieval
2. Deep Learning - Image Retrieval
3. Linear Regression - House Prices
4. Logistic Classifier - Sentiments
5. Recommender System - Songs

Data

Programming Language

Python

Packages

Anaconda, Graphlab Create Installation guide

Tools/IDE

Jupyter notebook (IPython)

How to use it

1. Fork this repository to have your own copy
2. Clone your copy on your local system
3. Install necessary packages

Note

This repository does not contain optimal machine learning models! It only assesses various models that can be built using different machine learning algorithms (either implemented or used directly from Graphlab Create package) to perform different tasks.