Machine Learning On Your Data (from Hacker News)

http://hn.premii.com/#/article/12279415

Top of hacker news today, light machine learning. Figured I would try it out on some of my spare data-collected and see if it spits out something interesting.

Will post graphs once I get it going.

2 Likes

vw-varinfo -k --loss_function squared eliot.train | tee scores.txt
FeatureName HashVal MinVal MaxVal Weight RelScore
^Sleep10-11hrs 237653 0.00 1.00 +0.2825 100.00%
^sleep_under4hrs 230339 0.00 1.00 +0.1051 37.21%
^vivid_dreams 96937 0.00 1.00 +0.1008 35.69%
^Steps10-12.5k 223453 0.00 1.00 +0.0806 28.54%
^Sleep_4-5hrs 261668 0.00 1.00 +0.0576 20.39%
^sleep11-12hrs 150232 0.00 1.00 +0.0510 18.04%
^sleep>12hrs 176848 0.00 1.00 +0.0507 17.96%
^Steps_0-2.5k 209658 0.00 1.00 +0.0418 14.79%
^Steps12.5-15k 165476 0.00 1.00 +0.0395 13.97%
^Sleep_5-6hrs 251919 0.00 1.00 +0.0376 13.32%
^write 129676 0.00 1.00 +0.0324 11.46%
^exercise 190069 0.00 1.00 +0.0238 8.42%
^vitamins 252959 0.00 1.00 +0.0093 3.30%
^shower 215555 0.00 1.00 +0.0050 1.76%
Constant 116060 0.00 0.00 -0.0047 0.00%
^Steps5-7.5k 92354 0.00 1.00 -0.0044 -1.56%
^steps>15k 199322 0.00 1.00 -0.0098 -3.46%
^Steps_2.5-5k 251927 0.00 1.00 -0.0182 -6.45%
^Sleep_8-9hrs 79384 0.00 1.00 -0.0307 -10.86%
^Sleep_7-8hrs 203880 0.00 1.00 -0.0354 -12.52%
^Steps7.5-10k 109514 0.00 1.00 -0.0436 -15.44%
^Sleep_6-7hrs 61052 0.00 1.00 -0.0497 -17.60%
^publish 12600 0.00 1.00 -0.0986 -34.89%
^Steps_above15k 31266 0.00 1.00 -0.1092 -38.65%
^sleep9-10hrs 51270 0.00 1.00 -0.1183 -41.88%
^Steps10-15k 24577 0.00 1.00 -0.1532 -54.25%
^diet 219124 0.00 1.00 -0.1573 -55.68%

which is to say that my 1/0 diet metric correlates with weight loss, but so does certain step and diet values.

This is almost exactly what I’m going to be doing… but I don’t have one built-in question… I’m just gathering qualitative data… to that end, I built an app using the hillariously easy-to-use MIT App Inventor… check out my project here: http://ai2.appinventor.mit.edu/#4930020062789632

Or you can install the apk directly here: tracker.apk (1.5 MB)
I’m accepting any/all feedback!