Jerry Ling
Jerry Ling
It's similar to why naive `for row in eachrow(df::DataFrame)` is slow, but if you use `DataFramesMeta.jl` or convert to a `TypedTable` from `TypedTables.jl` the row iteration would be faster again....
Interesting, I will take a look
hmm, I guess because people manipulate the modes, we need a more robust way to find the correct mode?
That's really weird, we haven't touched anything in a while, and I don't see any new release in XrootD.jl or the XRootD_cxxwrap_jll tagging others may know @peremato @giordano I'm trying...
so when I first saw your issue, I immediatedly tried in a new `]activate --temp` envoronment and I saw this: ``` Precompiling project... ✓ JLLWrappers ✗ UnROOT 1 dependency successfully...
wait, is XRootD wrapper just broken on Windows?
Right.... Well I think in that case XRootD.jl should have unit test on CI to make sure it doesn't fail to precompole and only throw an error message at runtime...
@TheBreadFarmer can you check if this solves the problem for you :eyes:
>Custom loss function how far are we from having this?
yeah I think having something like https://dmlc.github.io/XGBoost.jl/dev/features/#Setting-a-Custom-Objective-Function is a good start. For my application, I was initially looking at https://arxiv.org/abs/1806.00322 to try to use a different loss function but then...