Using chord as the base of melody mining is novel and interesting. The presentation is clear and informative. -- Professor Tian Zheng

Self designed the whole project flow, directed team of 5 to build an R Shiny app MelodySoup to generate chord progressions, designed logo. Conducted sequence analysis in R on chords of 990 songs using n-gram and tf-idf, visualized chord sequences by interactive chord diagrams and Sankey diagrams. Verified genre classifiablility by comparing prediction accuracy of Naïve Bayes and baseline model (improved by 107%). Introduced MelodySoup using ioSlides to class of 62 students, 2 professors, and 4 teaching assistants, scored highest marks in class

Listen to the music piece built based on a chord sequence recommendated by our app:

Stack Overflow is the largest online community for programmers to learn, share their knowledge, and advance their careers.

As future data scientists, we cannot live without Stack Overflow when programming. To help us better understand and use this community, in this project, we are going to answer the following 3 questions:

1. What can we know from comparing R questions with Python questions?
2. How can we ask questions that are more likely to receive acceptable answers?
3. Can we create a recommendation system to automatically generate tags for Stack Overflow questions?

In this project we will explore the association between music features and lyrics words from a subset of songs in the million song data. Created lyric words recommender algorithms in R for music pieces using music features. Extracted 1125 music features from 2350 songs of h5 format, cleaned lyric text data, converted word count table to binary table. Conducted topic modeling on binary lyric text, obtained 'topics_word' matrix of word distribution for 20 topics. Trained Random Forest model on music features to predict 'doc_topics' matrix of topic distribution for songs. Predicted word probability matrix for 100 testing songs by multiplying predicted 'doc_topics' matrix and 'topics_word' matrix, prediction performance ranked second in class

In this project, we try to propose a feasible improvement on the currant classification engine to classify poodle dogs and fried chicken in terms of running cost (storage, memory and time) and prediction accuracy. The baseline model uses gradient boosting machine (gbm) with decision stumps on 5000 SIFT features. Our advanced model uses ensemble method (majority vote) of adaboost, random forest, knn, xgboost, and gbm on combination of 521 RGB features and original 5000 SIFT features, which enhances the accuracy rate by 21%.

This project explores and visualizes the noise level in New York City by integrating analyses of the 311 complaints data in 2015 on NYC Open Data Portal, the geographical data of construction sites, fire stations, hospitals, and clubs in NYC. We created a Shiny App to help users navigate through our findings in 3 main tabs: Statistics, Map and Data.

Analyzed ACS Public Use Microdata Sample (PUMS) 2014 data (2.97GB with 3.13M entries) using R in team of 5. Investigated and visualized income gaps by Marriage Status, Motherhood, Location, Race, and Education within each occupation of women in R. Presented analysis results to class of 62 students, faculty, and 4 teaching assistants using R Markdown Notebook

Conducted research on news, experts’ opinions, facts, and figures to analyze China's current financial sector policies. Developed reform strategies on organizational restructuring, shadow banking monitoring system, and stock market transition. Wrote 2-page policy memo, pitched to a panel of 5 judges comprised of professors and field experts

This is the last final project of Advanced Data Science which we are allowed to do whatever we want. Self-designed whole project flow, directed team of 5 to build an R Shiny app MelodySoup to generate chord progressions, designed logo. Conducted sequence analysis in R on chords of 990 songs using n-gram and tf-idf, visualized chord sequences by interactive chord diagrams and Sankey diagrams. Verified genre classifiablility by comparing prediction accuracy of Naïve Bayes and baseline model (improved by 107%) Introduced MelodySoup using ioSlides to class of 62 students, 2 professors, and 4 teaching assistants, scored highest marks in class.

This is the last final project of Advanced Data Science which we are allowed to do whatever we want. Self-designed whole project flow, directed team of 5 to build an R Shiny app MelodySoup to generate chord progressions, designed logo. Conducted sequence analysis in R on chords of 990 songs using n-gram and tf-idf, visualized chord sequences by interactive chord diagrams and Sankey diagrams. Verified genre classifiablility by comparing prediction accuracy of Naïve Bayes and baseline model (improved by 107%) Introduced MelodySoup using ioSlides to class of 62 students, 2 professors, and 4 teaching assistants, scored highest marks in class.

This is the last final project of Advanced Data Science which we are allowed to do whatever we want. Self-designed whole project flow, directed team of 5 to build an R Shiny app MelodySoup to generate chord progressions, designed logo. Conducted sequence analysis in R on chords of 990 songs using n-gram and tf-idf, visualized chord sequences by interactive chord diagrams and Sankey diagrams. Verified genre classifiablility by comparing prediction accuracy of Naïve Bayes and baseline model (improved by 107%) Introduced MelodySoup using ioSlides to class of 62 students, 2 professors, and 4 teaching assistants, scored highest marks in class.