Projects

Tools: python - beautifulsoup4, pandas, seaborn.

Skills: web scraping, data visualization

Used car prices have gone through the roof with the pandemic and the war in Ukraine, which has made getting a decent replacement for my recently-totaled (by someone else) '06 Prius quite challenging.

I've spent a lot of time in the last couple of months on Cars.com looking for a deal that would neither break my bank nor have a high probability of becoming a headache in a year - to no avail. I did some research and it looks like prices might begin to decline sometime soon, although no one can tell for certain.

So I asked myself: How long should I wait? Should I buy a car that recently got listed or should I wait a bit before getting in touch, see if the price changes? Are certain models getting cheaper than others? To answer those questions I decided to build a Python script that runs every night while I'm asleep. It scrapes Cars.com for the filters I selected and adds the details of the cars it found onto a table in a csv file. From there, I can read the file in Jupyter Notebooks whenever I want and use Python to see what the prices of the cars I'm after are doing.

I don't have much data yet, but I expect this tool to become more powerful as time goes by. So far I have found that only around 10% of cars change price in a week, and cars with higher mileages tend to drop in price faster than those with lower mileages.

Want to see this project? Click on the title and it will send you to a Jupyter notebook showing my code and what I have done with it so far.

Tools: python - pandas, seaborn, tkinter.

Skills: building GUIs, data visualization.

For those who have never heard of it, 'Magic: The Gathering' (or MTG, for short) is the most played collectible cards game in the world, which begins to explains why my cousin loves it.

I mean, he spends hours a day looking through the twenty thousand unique cards on the market deciding which ones he will add to the next deck he's building. That does, as one might expect, come with some challenges.

Search tools are available to facilitate finding the right cards, but my cousin was having trouble with balancing out his decks. You see, MTG cards have a 'cost' that must be 'payed' by the player when they want to play it. If a deck only has high cost cards, it won't work properly. But the same goes for a deck that only has low cost cards. This is also true for other properties of the cards, such as their color, type, etc. And given that the decks my cousin builds are composed of 100 cards each, it becomes difficult to keep track.

This is where I saw an opportunity to help him out. I thought that it shouldn't be too hard write up a program that took in a list of cards in a deck and spit out a few useful graphs and numbers. My cousin is not very comfortable with tech, so I knew I would need a GUI. So I did some research and decided to use Tkinter, a Python package for creating GUIs in Python. I hooked up the GUI to a few functions, and using Pandas and Seaborn I was able to make it display graphs. My cousin has been using it, and while he has some suggestions for improvement, I think it's been a success!

Tools: github; python - pandas, mne, librosa, scipy, seaborn.

Skills: analyzing EEG data, signal processing, group collaboration.

One of the most interesting courses I've taken at UCSD - COGS 138: Neural Data Science - was also one of the hardest to grasp. While I have quite a bit of background in neuroscience and in data science, analyzing EEG data is not an easy task.

For our project, my group of 5 decided to use a dataset from a study that was made available online. The study was interested in using music to change participants' affects, and using their EEG data to reverse-engineer this into an algorithm that could read people's affects from their EEG readings. We used some of their data for a different purpose - we wanted to see if we could find correlations between the features of the music being played to participants and artifacts in their EEG readings.

This project was challenging for many reasons. For starters, we had trouble downloading the data onto a platform we could all access and work with simultaneously. After dealing with that, the data we wanted to work with was nested and stored across several tables. Linking the EEG readings to the corresponding song being played was also a challenge. Once we finally got through all of that, we had to deal with the largest challenge of them all - how do you extract meaning from EEG data?

Luckily, we had a tenured professor of Cognitive Science there to help us, Prof. Bradley Voytek. He gave us tips on using MNE, a Python package designed specifically for this purpose, as well as other tools he designed himself in his lab. At the end of the course, we did not find any statistically relevant results but learned a lot from trying to.