CI for master is currently failing. This PR makes changes to test code to fix this. It makes no changes to source code.
There are two tests that have become invalid, causing the fails:
A test dependency, MLJDecisionTreeInterface, made a breaking change (the hyperparameter pdf_smoothing was removed from DecisionTreeClassifier). Since, currently, test dependency versions are unrestricted in TreeParzen.jl, this change appeared unexpectedly. Moving forward, I suggest test dependencies be specified using [dep] entries in a separate test/Project.toml file (instead of the root Project.toml file). See this example. However, this will require julia compat to be bumped to at least Julia 1.2. I think Julia 1.6 and above is fine now for testing - Julia 1.6 is the current Long Term Release and many packages no longer support earlier versions.
Distributions.jl made a change in version 0.25.44 in the way that it throws errors for probability vectors that do not sum to one. Previously an ArgumentError was thrown. But now a DomainError is thrown instead. This was technically breaking but not tagged as such.
CI for master is currently failing. This PR makes changes to test code to fix this. It makes no changes to source code.
There are two tests that have become invalid, causing the fails:
A test dependency, MLJDecisionTreeInterface, made a breaking change (the hyperparameter
pdf_smoothing
was removed fromDecisionTreeClassifier
). Since, currently, test dependency versions are unrestricted in TreeParzen.jl, this change appeared unexpectedly. Moving forward, I suggest test dependencies be specified using [dep] entries in a separate test/Project.toml file (instead of the root Project.toml file). See this example. However, this will require julia compat to be bumped to at least Julia 1.2. I think Julia 1.6 and above is fine now for testing - Julia 1.6 is the current Long Term Release and many packages no longer support earlier versions.Distributions.jl made a change in version 0.25.44 in the way that it throws errors for probability vectors that do not sum to one. Previously an
ArgumentError
was thrown. But now aDomainError
is thrown instead. This was technically breaking but not tagged as such.