ashwanirajan Goto Github PK

a-meta-analysis-of-neutron-lifetime-measurements

We generated Error Distributions for the neutron lifetime using a subset of measurements compiled in the 2018 edition of Particle Data Group.PDG and checked the gaussianity of the error distribution using the weighted mean and median estimates.

a-scalable-song-recommendation-system

We extracted and preprocessed the One Million Song data, user profiles, and song lyrics data using Spark on Databricks and AWS EMR and stored aggregated features as MongoDB collections. We built collaborative and content-based song recommenders, popularity, and genre predictors using distributed models for Matrix Factorization, word2vec K-means clustering of song lyrics, and classification & regression models using Spark MLlib. We also developed a web app prototype to showcase the above results.

analysethis-amex

AnalyseThis 2019 Submission.

bitcoin-mining-pool-detection-and-identifying-anomalies

Bitcoin provides an extra level of anonymity for the identities of the users, since the user addresses are hashed. A Mining pool is the pooling of resources by miners, who share their processing power over a network, to split the reward equally, according to the amount of work they contributed to the probability of finding a block. I attempt to create a classifier model which identifies the mining pools from other users and then detects anomalies in these mining pools.

central-limit-theorem-usf-msds610

Class Project for code demonstration. Topic: Central Limit Theorem

clustering-of-wogma-movie-reviews

This repository includes python script to extract, download and cluster movie reviews on wogma.

crop_monitoring

Crop monitoring using satellite imagery

eda-and-visualization-hr-attrition-report

We performed Exploratory Data Analysis and did some cool visualizations to study the reasons for attrition in a given company.

experimentation-streaming-platform

fake-news-detector

FAKE NEWS DETECTOR Running the code : We can run the main.py file. You just need to input whether you need the prediction to be “binary” or “multiclass”. About the model: I have used the statement as well as the justification together as text input to train the model. I merged the statement and justification as a single text. First we use pre-trained 100 dimension Glove embeddings. This constitutes the first layer of the sequential model. Then we have used a 1D Convnet with relu activation, and filters =32. We added another layer of 1D convent with the same activation and filters = 64. Then we used Bidirectional LTSM with two layers(128, 64) and followed by a softmax layer, which has 2 or 6 neurons based on our input as “binary” or “multiclass”. Scope of improvement : Due to limited time, I could not attempt a few more possibilities. The accuracy can be improved by the following few measures: 1. Using the metadata related to previous counts or true/fake statements of each speaker in both binary and multiclass. 2. Using parallel BiLSTM as mentioned in paper might improve the results to a good extent. 3. Using both training data and validation data to train our final model will improve accuracy to a minimal extent as well. Accuracy : 1. Multiclass : 0.832 2. Binary : 0.523

fastai_projects

A compilation of all the projects I did while taking the online MOOC by USF on Fastai

hurricane-damage-detection

implicit-rating-predictions-on-booking.com-data

Experiment with Machine Learning Model that can predict whether a user will prefer a trip destination recommendation given the historical user-item interaction data.

investigating-clusters-in-pulsar-data

logisticmap-mandelbrotset

machine-learning-in-scikit-learn

Machine learning lab course (USF's MSDS 699)

myfastaijourney

nasa-spaceapps

national-taiwan-university-internship

My work during the internship at NTU, Taiwan.

nbody-simulations

non-gaussian-error-distributions-of-galactic-rotation-speed-measurements

We generated Error Distributions over a compilation of 162 galactic rotation speed measurements from literatureusing different central estimates using Python. We then compared the distributions to different standard models such as Gaussian, Cauchy, Double Exponential andStudent’s t using K-S tests and Chi-squared tests and inspected the reasons for their non- gaussianity using Python.

nptel-data-structures-and-algorithms-python

object-detection

Object Detection for vehicles on COCO Dataset

object-detection-openimages

Vehicle Detection using State-of-the-art Object detection and Localization techniques on Open Images Dataset

one-million-song-analysis

python-tutorials

This repository contains a tutorial for all the different methods for String Formatting available in Python.

shala2020.github.io

Lectures and notes for a summer course on DS, ML, and AI

simulating-an-epidemic

understanding-dark-matter

used-cars-price-prediction

We used Linear Regression Analysis to predict the prices for used cars.