Šimon Mandlík

Yes, there are still some problems. `!a | !b` doesn't work for me either

Yes, I'm using 0.6.1. The problem most probably is when vim-matchup is included. Are you able to reproduce with this init? ``` set nu lua

My only experience of large datasets and MLDataPattern was one usecase, when feature array couldn't fit into memory, however simple vector of indices could (dataset comprised around 10^10 examples). I...

Agreed. I believe that now when `RandomObs`/`RandomBatches` are available, it makes sense to have `BalancedBatches` in addition to `BalancedObs` as well.

Don't merge until there is a compatible version of `MLLabelUtils` as well.

PoC for deduplication in NGramMatrices ```julia julia> function Mill._mul(A::AbstractMatrix, S::PooledVector, n, b, m) C = zeros(eltype(A), size(A, 1), length(S)) iz = Mill._init_z(n, b) idcs = Dict(r => Queue{Int}() for r...

Sure, and deduplication of instances in `BagNodes` as well. That said, it is possible that in some cases the vanilla version will still be faster

One more idea is to use an array of bag lengths instead of `UnitRange`s. It will be slightly less convenient, but will save memory

Why not? For `ArrayNode` we basically propagate `vcat`, `hcat` and `catobs` to `data` and `metadata`.

That was it, thanks :) A dumb error when rewriting config from vimscript to lua