2. today
• First week of /cure/
– What happened?
• Next steps
– Changes for next deployment
3. Website
• 347 distinct site visitors
– 62% US, 29% China
• 38% of visits > 2 minutes
• 4.7% of visits > 30 minutes
• analytics
4. Players
• 125 registered, 108 played mammal,
• 58 finished a board on cancer challenge.
of the 58 players
Have Cancer Consider Most recent degree
Knowledge themselves a
Biologist
no 41% no 53% BS 29%
NA 1 NA 1 MA 17%
yes 55% yes 45% NA 3%, other 3%
PhD 47%
similar proportions if include the
ones that stopped at mammal
5. Registrations
36 registrations on
our Sunday, mostly
China created
9/14/12
9/13/12
9/12/12
9/11/12
9/10/12
created
9/9/12
9/8/12
9/7/12
Launch day
9/6/12
0 20 40 60 80 100 120 140
Accumulating registrations
7. Time spent
RyanM 23 hours
Shenyije
Stolo
90000
80000
70000
60000
50000
Total seconds 40000 Series1
between cards 30000
played within 20000
one board 10000
0
0 10 20 30 40 50 60
Player
About 61 hours spent in total.
This is a lower bound.
10. So is it working?
• Baseline for random gene selection within our
2500 genes.
– Random samples of 10 genes, n = 1000
– cross-validation performance
– min: 47.46922024623803
– max: 67.44186046511628
–mean: 55.12311901504788
– std dev: 4.140814369975244
– median: 54.651162790697676
11. Gene ranking take 1
• For each board
– find the first hand played by each (unfiltered)
player
• for all of those hands (min 5 diff players)
• calculate the frequency of selection F for each gene.
• Sort genes according decreasing F
12. Ehh, sort of maybe
• Frequency-based selection, all players, top 10
ranked genes
– CV = 60
• Frequency-based selection, all players that
declare knowledge of cancer and have played
< 13 cards per hand on average, top 10 ranked
genes
– CV = 63
13. Next steps
• Open – how to stop people from optimizing
hands based on cv performance yet still give
them feedback on what is generally working?
• How to make it more fun?
• How to collect more useful data?
Editor's Notes
for 25 genes mean is 56. otherwise almost exactly the same