Percentage sum over training, validation and test datasets (EMiner)

CarlosP73 — Sat, 18 Apr 2020 21:48:33 GMT

Hi all,

I always wandered (but actually never checked until now) how EMiner (working with v15.1) deals with the percentages for tr, val and test ds, as you're allowed to enter numbers from 0 to 100 without any restriction at all.

I just tried today and entered 45%, 35% and 25% (adding up 105%) and surprisingly does not issue any error or warning

In order to finish my little test, I opened the number of records for each and below is what each ds actually got:

Tr(45%) -> 42.84%

Val(35) -> 33.32%

Test(25) -> 23.83%

Without any further calculation I just reckon it spreads the excess uniformly... but haven't found any documentation whatsoever...

Does anyone know what actually happens behind the scenes?

Thank you!

Carlos

Re: Percentage sum over training, validation and test datasets (EMiner)

WendyCzika — Mon, 20 Apr 2020 15:27:02 GMT

It's dividing those values by 105 instead of 100 to get the percentages.

42.8% = 45/105 ...

Re: Percentage sum over training, validation and test datasets (EMiner)

CarlosP73 — Mon, 20 Apr 2020 15:58:20 GMT

Makes complete sense... Thank you very much Wendy!

topic Re: Percentage sum over training, validation and test datasets (EMiner) in SAS Data Science

Percentage sum over training, validation and test datasets (EMiner)

Re: Percentage sum over training, validation and test datasets (EMiner)

Re: Percentage sum over training, validation and test datasets (EMiner)