CONSISTENCY: lambda penalty is not on the expected scale in get.Zt

pneuvial / c3co

Inferring cancer cell clonality from copy-number data

5 stars 1 forks source link

CONSISTENCY: lambda penalty is not on the expected scale in get.Zt #57

Closed jchiquet closed 5 years ago

jchiquet commented 5 years ago

The glmnet is solving 1/2 RSS/nobs + λpenalty. We use it to solve something like RSS + λpenalty. So to be consistent with the value of lambda provided by the user, we should divide by a factor (2 nobs), which is (2 n * J) in the case at hand.

HenrikBengtsson commented 5 years ago

Note to ourselves: I just wanted make sure we're not adjusting lambda prior to calling get.Zt(); I just checked positiveFusedLasso() and I can confirm that lambda is not changed within that algorithm.