Ryan Peach

This code implements the Hungarian (Munkres') and Murty's algorithms in raw Python 2 and NumPy. It is written purely functionally to be easily memorized and parallelized. It is also based on research sources to be used in artificial intelligence programs that require optimal decision-making where an assignment is one of the tasks.

It is ready to run just by calling "python assignment.py." Given a "NxN" matrix, this algorithm minimizes the sum of the values giving a unique assignment of one row per column and returning a length N vector containing the column index for each row assignment. Murty's algorithm does this multiple times in ascending order of cost by using a Python generator. This "O(n!)" algorithm is implemented in an "O(N^3)" time complexity using the latest research.

In Knowledge-Based AI, an intensely project-and-research-based graduate course, we each created our own research level artificial intelligence—each of which should be capable of solving every standard set of the Raven's Progressive Matrices intelligence test, verbally at first, visually by the end of the class, and were graded on our agent's performance on these tests.

My final agent scored 19/24 on the hardest problem sets, purely via visual information. My first project reflection was selected as a top 10 best in class, and my agent overall scored very well relative to my peers. We were allowed no outside code from other students, and worked with Python using only Numpy and PIL libraries. All in all, this has been my favorite class by far overall, and it inspired much of my interests in AI going forward.

While working for PNNL, I was given the task of constructing a mathematical model to increase the detection of Twitter-traffic spikes in a large web-traffic time-series. Over the course of several months, working with a mathematician Ph.D. and 2 undergraduate students of differing specialties, I learned how to write fluently in Python—using several different mathematical and graphical libraries and wrote thousands of lines of well-documented code for every task imaginable. My primary line of research was in the area of Fast Fourier Transforms (FFT), which required a lot of rigorous mathematical study and code manipulation on my part.

I spent months writing and testing highly complex algorithms for mathematical analyses, and even more time on interpreting the data and trying to improve the graphical outputs my code would generate. In the end, the experience was highly rewarding, and it taught me how to learn and research independently, as well as how to function well in a business environment.

In my spare time, I work on reinforcement learning problems on OpenAI Gym.

The following is a sample of a Q Learner with documentation that I have recently written for the site; which has a very high ranking and quick learning time for the environment, and is highly generalizable to other problem sets.

My entire GitHub profile is available online.

I have participated in leading edge open-source projects such as OpenNARS at Temple University and frequently submitted code snippets to scientific computing archives such as NumPy and SciPy.

I upgraded their development tools to the most recent Sklearn standards and constantly answer issues that arise.

I did some work on data analysis after the recent election and received some much-appreciated acclaim for my post on Kaggle for using their datasets.