Closed rikhuijzer closed 2 years ago
I'm personally against this. The PR #121 you made for the internal docstring makes sense to me but I don't think we should go beyond that. The naming of symbol here goes with the literature I was working with and helps uniformity (it's fairly standard though, of course, there may be other schemes out there).
We can always make the docstring for the constructors more legible but given it's extremely common to have lambda used for both the l1 and l2 problem, I don't see a problem doing the same when there's a single regulariser so that it's clear what it affects.
Currently
lambda
is the strength ofL1
if:l1
,L2
if:l2
andL2
in:en
which is very tricky. It is easy to mistake them when switching thepenalty
. Should we refactor this to