Rao Hamza Ali

Converting Piano Compositions into Paintings

Using historical contexts, we implemented a rule-based system that converts piano compositions into paintings, taking key musical features into account. Work is underway for the conversion in the other direction.

Used Python, noise fields, clustering algorithms, and the Processing framework.

Read the Publication

Delineating Neighborhoods from Geocoded Appraisal Data

Using only the distance between subject and comparable appraised properties, we generated neighborhood boundaries that explain more variation in properties than ZIP Codes or Census tracts.

Used Python, R, Hierarchical Distance Based Spatial Clustering with Application to Noise (HDBSCAN), custom designed spatial filters, regression models, and clustering algorithms.

Read the Publication

Application of Survival Analysis to Open Source Software Repositories

We applied statistical survival analysis to open source Python software repositories to identify the key issues that affect the health of repositories on a long run. Having multiple hosting services, a continuous timeline of publishing releases, and a good network of developers, positively affect a repository's health.

Used Python, R, Kaplan-Meier curves, Cox-Proportional Hazards model, and censoring.

Read the Publication

Modeling Exhausted Topics for Programming Languages on Stack Overflow

We generated topics on over 5 million questions asked on Stack Overflow regarding programming languages, to identify the topics that are no longer being questioned on. We found distinct topics for Python, JavaScript, Java, C++, and R that users on Stack Overflow no longer ask questions about as answers to those questions may already be present on the website.

Used Python, R, topic modeling, Latent Dirichlet Allocation, gensim, spaCy.

Read the Publication

Identifying Astronauts in Photos Taken Inside the International Space Station

In collaboration with space archaelogists, we created a system to automatically identify astronauts in photographs of the crew inside the space station. We first attempted to create our own ensemble of deep neural networks for the identification process but ultimately used Amazon's Rekognition API for a working solution. The system was tested on unpublished photos provided by NASA and showed a 92% match rate for crew faces. We further made inferences about the social structure on the space station using the match results.

Used Python, R, TensorFlow, Keras, VGG19, Amazon's Rekognition API, and graph algorithms.

Publication Under Review

Using Neural Embeddings to Represent Artists' Art Periods

Through the application of word embeddings models on paintings of famous painters, we generate fingerprints of artworks that represent unique art periods. The approach involves generating sliding windows for paintings to extract color values and generating a bag-of-words model for the colors.

Used Python, gensim, word2vec model, and the Processing framework.

Publication Under Review

Studying the Sentiments Around the 2020 US Elections on Twitter

We applied sentiment analysis to a corpus of 7 million tweets sent out during the 10 days around the US Election Day. We took a novel approach to also view the sentiments of deleted tweets and user accounts, and incorporated that into our overview of sentiments for the presidential candidates. We also make a recommendation for future studies to use deleted social media posts into account when generating sentiments around an event.

Used Python, R, and the VADER sentiment analysis tool.

Publication Under Review