Closed ablaom closed 2 years ago
I'm sorry, the docs currently only reflect the master branches of the corresponding packages. Installing both OutlierDetection
and OutlierDetectionData
from master should work, only ODDS.read
was changed to ODDS.load
in OutlierDetectionData
.
There's still an issue with fit
, unless one also has MLJBase (or MLJ) loaded. The source of this is your new use of MLJModelInterface.matrix
. And you have left an adjoint behind, which does not make sense now X
is a data frame.
Maybe you want to avoid MLJ altogether in your "local" API. I don't see anything wrong with using matrices there, but if you want to allow generic tables input, then you could use Tables.matrix
instead of the MLJModelInterface one, which does not work unless MLJBase is loaded.
Of course, if you don't care to have a separate local API at all, then these comments are not relevant.
Good point, I didn't know that MLJModelInterface.matrix
requires MLJ. I changed to the Tables.matrix
-conversion for now, but might switch to matrix-only in the future.
@davnn I wonder if it wouldn't be timely to make some sort of doc update, now that the detector models are MLJ-discoverable and the API has stabilised somewhat. It needn't too comprehensive just now, but correct would be good 😉
@davnn I wonder if it wouldn't be timely to make some sort of doc update, now that the detector models are MLJ-discoverable and the API has stabilised somewhat. It needn't too comprehensive just now, but correct would be good 😉
I'm working on it 👍
Edit: One reason why the docs are not finished yet is that I'm not happy with the API as it is right now, see https://github.com/alan-turing-institute/MLJ.jl/issues/780.
The README and documentation is up to date now.
EDIT: @ablaom MLJ docs will follow shortly.
Looks great, thanks!
I guess
X[train,:]'
was meant here?And
score
does not appear to be defined: