Intro Data Science/Kaggle update

The semester is over, so here’s a little update about the Intro to Data Science class (previous post).

Kaggle Final Project

The final project was a Kaggle competition to predict standardized test essay grades. Although I still had lots of ideas, when I wrapped up a week early I was in first on the public leaderboard, and maintained that to the end. After it was over the private results gave first to Maura, who implemented some awesome ensembling. For commentary take a look at Rachel’s blog post. There’s a bit of discussion in the forum, including my write up of my code.

Visualization

During the competition I maintained a visualization of the leaderboard, which shows everyone’s best scores at that moment. Will Cukierski at Kaggle appreciated it, and apparently the collective impetus of Rachel and I encouraged them to make a competition out of visualizing the leaderboard! See Rachel’s blog post about it for some more info (and a nice write up about my mistakes).

Now back to studying for finals…

Intro to Data Science – Kaggle Leaderboard

This semester I’m auditing Rachel Schutt’s Intro to Data Science class. I originally registered for it, but at the end of the add/drop period decided I wasn’t confident in my academic background, and wasn’t sure about the workload that would be required. In retrospect it was a mistake to drop it. However I have been attending class as I can (about half the time).

The final project, accounting for most of the grade, is a Kaggle competition. It’s based on an earlier competition, and the goal is to develop a model to grade standardized test essays (approximately middle school level). As an auditor Rachel asked me not to submit, but my cross-validation suggests my model (linear regressions with some neat NLP derived features) is still besting the public leaderboard (Quadratic Weighted Kappa Error Measure of .75), but who knows.

I thought I could easily adjust my MITRE competition leaderboard graph to Kaggle’s CSV, and it turned out to be pretty easy. The biggest issue ended up being that MITRE scored 0 to 100, and this scores 0 to 1. That had some unintended consequences. launchd + python + R should upload this every hour or so (when my laptop is running).

I’m frankly surprised Kaggle hasn’t done something like this before. Maybe if I have a bored evening I’ll try to do it in D3, which should look much nicer.

Update, 12/16: I’ve posted a followup after the end of the semester.

Births by Day of Year

Andrew Gelman has posted twice about certains days being more or less common for births, and lamented the lack of a good, simple visualization showing all 366 days.

Well, I heard his call! My goal was simply a line graph that showed

Finding a decent public dataset proved surprisingly hard – very few people with large datasets seem willing to release full date of birth (at least for recent data). I considered voter files, but I think the data quality issues would be severe, and might present unknown bias. There’s some data from the CDC’s National Vital Statistics System, but it either only contains year and month, or isn’t available in an especially easy to use format. There’s some older data that seemed like the best bet, and which others have used before.

A bit more searching revealed that Google’s BigQuery coincidentally loads the NVSS data as one of their sample datasets. A quick query in their browser tool and export to CSV and I had the data I wanted. NVSS/google seems to include only the day of the month for 1/1/1969 through 12/31/1988. More recent data just includes year and month.

SELECT MONTH, DAY, SUM(record_weight)
FROM [publicdata:samples.natality]
WHERE DAY >= 1 AND DAY <= 31
GROUP BY MONTH, DAY
ORDER BY MONTH, DAY

Some basic manipulation (including multiplying 2/29 by 4 per Gelman’s suggestion) and a bit of time to remember all of R’s fancy graphing features yielded this script and this graph:

See update at bottom!

I’ve labeled outliers > 2.3 standard deviations from the loess curve (which unfortunately I should really predict “wrapping” around New Years…), as well as Valentine’s and Halloween. You can see by far the largest peaks and valleys are July 4th, Christmas, and just before/after New Years while Valentine’s and Halloween barely register as blips.

It’s possible there data collection issues causing some of this – perhaps births that occurred on July 4th were recorded over the following few days? The whole thing is surprisingly less uniform than I expected.

Simulating Birthday Problem

I also wanted to simulate the birthday problem using these real values, instead of the basic assumption of 1/365th per day. In particular I DON’T multiply Feb 29th by 4, so it accurately reflects the distribution in a random population. This is data for 1969 to 1988, but I haven’t investigated whether there’s a day of week skew by selecting this specific interval as opposed to others, this is just the maximal range.

I did a basic simulation of 30,000 trials for each group size from 0 to 75. It works out very close to the synthetic/theoretical, as you can see in this graph (red is theoretical, black is real data). Of note, the real data seems to average about 0.15% more likely than the synthetic for groups of size 10-30 (the actual slope).

I’ve also uploaded a graph of the P(Match using Real) – P(Match using Synthetic).

If you’re curious about the raw results, here’s the most exciting part:

Update

Gelman commented on the graph and had some constructive feedback. I made a few cosmetic changes in response: rescaled so it’s relative to the mean, removing the trend line, and switching it to 14 months (tacking December onto the beginning, and January onto the end). Updated graph:

MITRE Challenge Graph

For my own curiosity I created a python + R script to grab the MITRE leaderboard and graph it. It’s a bit of python to grab the leaderboard and write out some CSVs. Then a bit of R code (updated link: http://a.libpa.st/4KFGq) generates the graph. It’s running automatically with launchd on my laptop, and it should be regularly uploading a png to the address below. Launchd is pretty awesome, but a royal pain in the ass to get set up. It doesn’t feel very deterministic.

I still need to figure out how to jitter the names so they don’t overlap (like YouGov & Agent Smith), but other than that I thought it was a nifty little exercise.

Each line is a team, with their best MAP scores as datapoints

Programming Challenges

I’m a fan of puzzles, programming and learning, so I’ve always enjoyed The Python Challenges. Recently my coworkers Delia, Chris and I came up with the idea of doing some of those within the company to help ourselves and our coworkers become more familiar with Python and R (and to a lesser extent SQL and other languages).

The end result is the YG Challenge, where we’ll be posting a few problems a week in at least R & Python, then solving them. Week 1 is up, and we have some great ideas for the future. Intended for our coworkers, it’s public because why not! Feel free to take a stab at solving them, especially if you haven’t used either of those languages before.