Clement Gendler

Chess Outcome Prediction

I examine a data set of 20,000 chess games and try to predict the outcome using a neural network and only the moves played. The moves are in a PGN format, so work needs to be done in order to prepare them for modeling. Data cleaning and preprocessing sections of the project handle that by putting the moves into split lists, putting those lists into an array, then tokenizing all unique moves. In addition, exploratory data analysis is conducted in order to gain a better understanding of the data and look into potential concepts to include into the model. Examples of EDA include looking at the distribution of game outcomes, a histogram of average ratings, and examining the most popular chess openings used by all and highly skilled players. I used LSTM and GRU neural networks to predict the outcome with an 88% accuracy for my best model.

Predicting Future COVID-19 Hotspots

I worked alongside two Data Science Fellows to create a model and map of future COVID-19 hotspots. Our study utilized COVID-19 daily data on confimed cases, population, weather, and COVID-related search activity to create our model. The breadth of the data (over 6 months of daily data for 50 US states) introduced a not-inconsequential amount of complexity. After testing SARIMA and ARIMA models, we tried two additional models: a Facebook Prophet and an IID Model. Ultimately, we determined that the IID Model yielded the most specific predictions, as it allowed us to fit models specific to each state. Our documentation includes heatmaps which visualize for the reader the difference in predictions.

Natural Language Processing Reddit Classification

The mission of this project is to scrape 5,000 posts overall from the anime and manga subreddits and use two models to predict from which subreddit the posts came based on the title. First, a function is created to retrieve all the data. Then, the data set is manually cleaned to ensure as high accuracy as possible and care when handling the unique range in post characters within these subreddits. Exploratory data analysis and preprocessing are handled via a WordCloud visual to show the word range, indicating prevalent words that can be attributed to one of the two subreddits. In addition, a plot of the top 10 words and their frequencies is displayed. RegExpTokenizer is utilized to manually tokenize title and post columns. Then, tokens are lemmatized to reduce verbs and improve accuracy even further. Preprocessing and lemmatizing functions were modified and mapped to the columns, looping back to join the tokenized data into a string form. Two models were tested: a basic LogisticRegression and a basic MultinomialNB, both with a CountVectorizer to further clean and process the data. To my surprise, the LogisticRegression proved to be more accurate, which I justify later within the project. I follow up by making conclusions and recommendations based on my findings.

Data Analyst

Built SQL queries in Snowflake, Google BigQuery, and Spark to combine various data sets for ad hoc and custom analyses. Used Looker to create dashboards and models within LookML, automating reporting to increase reports sent by 500%. Provided data-driven reports across DeepIntent’s product suite and client base, generating $1 million in incremental spend. Audited and identified improvement opportunities in our Patient Modeled Audiences, increasing overall reach KPI by 10%. Ran A/B tests to determine marketing technique efficacies using Python, SQL, and Excel (PivotTable, VLOOKUP, etc.). Developed innovative methodologies to derive, evaluate, and present performance benchmarks to clients. Troubleshot data mining, integration, and quality issues while owning campaign-related systems and databases. Created and maintained thorough documentation as the point of contact for dashboard report management.

Marketing Data Analyst, Crossix Analytics

Produced and managed DIFA HCP Site, a platform for informing, measuring, and optimizing healthcare marketing. Used SQL, Python, Excel, Spark, and data visualization tools to produce monthly deliverables that track web analytics. Analyzed root causes and data anomalies while monitoring and troubleshooting analysis execution. Automated, planned, and improved internal processes through collaboration with the Engineering team. Handled client inquiries related to data and DIFA methodology by providing insight to the Client Services team

Data Science Fellow

Completed 480-hour immersive course in python programming and data science . Solved real-world problems by applying data collection, cleaning, visualization, analysis, modeling, as well as machine learning and deep learning techniques to large data sets. Predicted a Chess game’s outcome using only moves played with LSTM and GRU neural networks. Analyzed and classified 5,000 Reddit posts using Natural Language Processing. Forecasted future COVID-19 hotspots with Prophet and IID models in collaboration with two Data Science Fellows

Marketing Analyst Intern

Analyzed qualitative research data for a mattress and personal care company via Excel. Developed summary analysis for mattress research report, providing companies with insight into consumer choice and gaining an advantage over competing companies. Designed a research presentation via PowerPoint, placing emphasis on visualizing summary analysis with a diverse variety of charts to illustrate findings

General Assembly

480-hour immersive course in python programming and data science.

New York University

NYU Buenos Aires

Studied abroad at NYU