I do apologise for being very unclear.

Reflecting on my thread it looks lazy and disrespectful of forum protocol.

Ive been studying horse racing and other sport events with inputs ive put together.

I have many samples of input variables for individual horses,irrespective of horse field sizes.

Desired outputs being 1 for 1st, 2 for 2nd etc.Ive normalized the inputs and ran it through PCA.

Im not getting results i'd hoped for.Noisy data also cleansed to an extent.

So i would like to delve into individual race samples and have binary outputs for every paired combo of the horses to interprate the winning horses without having to manually do this.

Horse1 output 1st - 1

horse 2 output 4th - 0

etc

I thought having thousands of horses would have rectified the race field bias.and the lenghts beaten offset.