« first day (3824 days earlier)      last day (1191 days later) » 

BND
BND
7:25 AM
(1) what a fingerprint is & how it's represented in the model?

It’s a simply binary vector of 0s and 1s [0,1,1,0,0,0 ….]
The length is usually 2048, but can be 512 or more.
(2) how the model's being fit - ordinary least squares or maximum-likelihood estimation procedures are scale-equivariant.
For model fitting different regression models from sklearn: ort KernelRidge
Ridge, LinearRegression, SVR, KNeighborsRegressor, GaussianProcessRegressor, MLPRegressor, GradientBoostingRegressor, RandomForestRegressor but also TPOTRegressor (http://epistasislab.github.io/tpot/using/) which may have ot…
 
 
2 hours later…
Stephan Kolassa
Stephan Kolassa
9:36 AM
@andi: the question you link to is actually interesting. It really boils down to "how do I simulate a regression model with prespecified standard regression coefficients?" I think the low response is less due to people finding the question uninteresting, and more to the fact that it's long, and people don't like slogging through code. I upvoted it and will edit it with a TL;DR, I'd be interested in an answer there, too - to be honest, I got nowhere, though it shouldn't be too hard.
 
 
10 hours later…
andi
andi
7:40 PM
Thanks @StephanKolassa
 
 
3 hours later…
whuber
whuber
10:47 PM
@andi It's a fascinating question. (I hadn't come across it before.) I posted one fairly general solution with a sketch of of its justification and working code to experiment with it. cc @StephanKolassa
 

« first day (3824 days earlier)      last day (1191 days later) »