Neutralization output in [-5, 5] range?


I’m trying to use the available at GitHub - numerai/example-scripts: The official example scripts for the Numerai Data Science Tournament. However, what I see is that the neutralize function outputs is in range [-5, 5] - is this to be expected? or is there a bug in the example code? If it is to be expected then is there some additional processing applicable before using the output of neutralization in place of predictions which are expected in [0, 1] range? What am I missing?

E.g. when running describe on the prediction column in this line I am getting:
count 686769.000000
mean 0.006763
std 1.000001
min -4.321063
25% -0.671214
50% 0.004776
75% 0.679364
max 4.316229
Name: preds_model_target, dtype: float64

Thank you,

That is expected behavior from the neutralize function. You can see that the neutralized predictions are ranked to be in [0, 1] here

1 Like

Thank you so much for this reply - it made my day!

I thought the line you pointed to has something to do with ensembling and didn’t even consider it to be related.

Have a great day!