Rankpos > 5000 giving higher returns than Rankpos < 100

For some reason I can't rank things normally where I choose the top 100. The returns only look reasonable if I set it to rankpos > 5000

image

image

image

image

Hmm for whatever reason I have to rank it in reverse, then this model works. Puzzling.

The model seems to only be able to predict bad returns well. The Results pages bear out that fact. Additionally, if you click on the ⋮ menu on kerasnn II on the Validation Models page, predictions can be downloaded for advanced analysis purposes.

Hmm yeah that does seem to be the case.

Oh cool, didn't realize you could download that - thanks!!