use Scratch.jl

[jbrea]

Addresses #20.
Thanks @maleadt. I didn't know about Scratch.jl. I think this is much cleaner now.

[codecov-commenter]

Codecov Report

Merging #21 (5bae5d7) into dev (ec9cd80) will decrease coverage by 6.76%.
The diff coverage is 100.00%.

❗ Current head 5bae5d7 differs from pull request most recent head fd0a277. Consider uploading reports for the commit fd0a277 to get more accurate results

@@            Coverage Diff             @@
##              dev      #21      +/-   ##
==========================================
- Coverage   30.37%   23.61%   -6.77%     
==========================================
  Files           1        2       +1     
  Lines          79       72       -7     
==========================================
- Hits           24       17       -7     
  Misses         55       55              

Impacted Files	Coverage Δ
src/OpenML.jl	100.00% <100.00%> (ø)
src/data.jl	21.42% <100.00%> (-8.96%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[maleadt]

Yeah, that's better. Is there a reason you don't just encode the datasets as a static Artifacts.toml (you can set them to lazy to avoid downloading them all at package installation time)?

[jbrea]

OpenML has nearly 5k datasets right now and new datasets are constantly added. I prefer the Scratch.jl approach over having to update an Artifact.toml every time when new datasets are added.

[maleadt]

Makes sense! FWIW, the advantage of artifacts is that they are easily cached, while scratchspaces are more ephemeral.

[jbrea]

@ablaom I think this is ready to be merged.