tbarrow Goto Github PK

a-painterly-approach-color-in-deep-learning

atlas-internship-reports

[Link to SP2019 Presentation : https://drive.google.com/file/d/1J4E2KB0QgCRwZRyVcXk8hHNX3w0YonpU/view?usp=sharing] A collection of the reports and presentations I wrote during my three semesters of internships as a data science intern for the UIUC ATLAS Internship Program. These reports and presentations cover analysis of the apparent age of Fortune 500 CEOs, determined by facial recognition software, alongside their actual ages. Visualization and exploration of this data was done via Tableau

capstone-1-analytic-report-research-proposal

The intent of this project was to thoroughly describe and explore a dataset of choice, which I used a dataset of paintings containing relevant information such as year made, style, genre, and dimensions in pixels. Following this, I created three analytics questions to help explore the data, and answered them through the help of data visualization and narrative statistics. I then proposed further research that could be done with this dataset, which is relevant to both classification and machine learning.

capstone-2-narrative-analytics-experimentation

The intent of this project was to thoroughly explore a dataset through the help of visualization and then propose and outline an experiment plan including a hypothesis, rollout plan, and evaluation plan. The dataset I used was one covering the air pollution of different types of gases caused by pollution from 2000 - 2016 in Indianapolis. https://www.kaggle.com/sogun3/uspollution

capstone-3-supervised-learning

This project was performed using the dataset linked below. The intent of this project was to explore the data as thoroughly as possible, create a research question, model the variables and outcome of interest, and create the most efficient classifier possible for the dataset through inclusion and exclusion of different categorical variables. https://www.kaggle.com/uciml/forest-cover-type-dataset

capstone-4-unsupervised-learning

The intent of this project was to explore the usability and versatility of unsupervised learning for a dataset of my choice. I decided to use a dataset made up of 20,000 pieces of data corresponding to various patterns taken from 800 images of the Avila Bible. The dataset contained a column with twelve values, each pertaining to one of twelve assumed-to-be-present copyists within the bible, labeling which worked on which pattern. I removed this column from my dataset for the sake of unsupervised learning, yet used it to determine the accuracy of my KMeans algorithm for n_cluster values ranging from 8 - 15 through the use of ARI and Silhouette Scores, and additionally for cluster analysis at n_clusters = 12 via a contingency table. I additionally used multiple dimensionality reduction methods to explore just how many clusters we could find visually within our data, fine-tuning the parameters of each for the best results. My overarching research question for this project was, "Just how much will the presence of many outliers in all of our features influence how comparable our clusters are to the actual classes of our data?" As such, the second half of this project required me to individually winsorize each column in our dataset, and then reapply each technique to see what had changed from the original dataset. Since the Jupyter Notebook for this project was much more than 25MB in size, I've provided a link to a Google Drive with it: https://drive.google.com/open?id=1piDerwXTAoWUQyVM5C-pOK8PSIt3BuF5

final-capstone-project

Constructed a model employing NLP and supervised learning techniques to take in a House Resolution and return a product that condenses and summarizes key points, topics, and stances within the Resolution such that it maintains as much information as possible while allowing further accessibility to the layman

tbarrow.github.io

My portfolio website.