About Me

Hi my name is Truman Daniels. I’m a software engineer from Portland, Oregon captivated by the mechanisms that structure games and the way those mechanics frame our understanding of how a game works. I use machine learning to predict the outcomes of games with a model called Predictive Outcome Analysis (POA). The goal of POA is to expand the mastery behind games by maximizing opportunity in MMA using data analytics.

This image has an empty alt attribute; its file name is 20220330_165346-edited.jpg

I grew up playing strategy games, learning to read while playing Magic the Gathering with my brother and how to count playing poker with my mom. In school, I played sports (mostly basketball) but was frustrated by the presentation and understanding of in-game statistics by coaches and media. My relationship with Mixed Martial Arts first came from a curiosity with the technical aspects of fighting. Further research made me enamored by the possibilities of implementing data analytics in a game that has such a variety of techniques and strategy its players can employ.

Some of the skills I’ve acquired throughout this process:

  • Programming in Python 3
  • Working with the Command Line / Linux Terminal to execute programs
  • Data manipulation, cleaning and wrangling using Pandas including merging and joining dataframes. Creating and connecting to an SQLite database and performing SQL commands to retrieve data.
  • Applying dimensionality reduction techniques like SVD and PCA to reduce noise in sparse data and implementing feature embeddings using tensorflow
  • Visual data analysis and creating easy-to-understand graphical representations using Plotly, MatPlotLib and Seaborn
  • Data scraping and web crawling using Requests + BeautifulSoup and Selenium

Implementing machine learning models with the following libraries:

  • Neural Networks using Keras / Tensorflow and Pytorch
  • Gradient Boosted Decision Trees using Catboost
  • Linear Regression using Scikit Learn with emphasis on feature selection for causality and interpretability using parameter regularization methods like Lasso and Ridge regression
%d bloggers like this: