scaling

[brendano]

[AbeHandler]

scaling issue. not bug. good argument for rethinking the scale (not saying we necessarily need to worry about it, at least in the next month)

[brendano]

the solutions we briefly discussed included

• have gap b/w 0 and 1, then linear above 1
• sqrt scaling
• log scaling

[AbeHandler]

one comment, having tried out sqrt scaling.... Sqrt scaling squishes big counts more than small counts -- so small counts seem bigger on the overall Y scale. Advantage v. linear scaling: you can see small counts more easily. Disadvantage v. linear scaling: linear scaling tends to hide small counts, which gives a sort of visual smoothing effect. W/ linear scaling: it's easier to see "events" (bunches of big counts all together). This is probably turkable, altho perhaps not for this paper.

[brendano]

ah interesting. yeah could try the whole "Tukey's ladder" of transformations. next paper or two away

On Sat, Jul 30, 2016 at 11:18 AM, Abe Handler notifications@github.com wrote:

one comment, having tried out sqrt scaling.... Sqrt scaling squishes big counts more than small counts -- so small counts seem bigger on the overall Y scale. Advantage v. linear scaling: you can see small counts more easily. Disadvantage v. linear scaling: linear scaling tends to hide small counts, which gives a sort of visual smoothing effect. W/ linear scaling: it's easier to see "events" (bunches of big counts all together). This is probably turkable, altho perhaps not for this paper.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/AbeHandler/rookie/issues/223#issuecomment-236370736, or mute the thread https://github.com/notifications/unsubscribe-auth/AAAGjWk2yAiFxC7i2kyt_A0vurBivS-xks5qa2tMgaJpZM4JTB6O .

[AbeHandler]

linear

sqrt

[AbeHandler]

I actually think linear scale is better -- at least for Rookie newbies. it makes things more event-y. you know where to look. 8/3/16 user studies.