MLUtils v0.4.8

Diff since v0.4.7

Merged pull requests:

• Fix regression with building empty dataloaders (#203) (@ChrisRackauckas)

Closed issues:

• Port collated batch view from DataLoaders.jl (#29)

MLUtils v0.4.7

Diff since v0.4.6

Merged pull requests:

• add some guides (#200) (@CarloLucibello)
• depend on MLCore.jl (#201) (@CarloLucibello)
• add obsdim argument to obsview (#202) (@CarloLucibello)

Closed issues:

• Status of MLLabelUtils porting (#16)
• Reexport OneHotArrays.jl (#77)
• mapobs problems with composite datasets and vector indexes (#110)
• DataLoaders(..., parallel=true) hanging (#132)
• MLUtils seems quite heavy (#155)
• Flux CI failures caused by recent DataLoader changes (#198)
• slidingwindow should have a dims keyword (#199)

MLUtils v0.4.6

Diff since v0.4.5

Merged pull requests:

• More efficient ntuple generation in unsqueeze w/ Val (#178) (@BioTurboNick)
• support for collate function and for mapobs in DataLoader (#181) (@CarloLucibello)
• update workflows (#182) (@CarloLucibello)
• implement trues_like, falses_like + vector indexing for joinobs (#184) (@CarloLucibello)
• add make_moons generator + some doc reorg (#185) (@CarloLucibello)
• add rng argument to splitobs, undersample, oversample (#186) (@CarloLucibello)
• improve splitobs docstring (#187) (@CarloLucibello)
• mention obsview in splitobs docstring (#188) (@CarloLucibello)
• fix batchseq for vectors of vectors (#189) (@CarloLucibello)
• fix doctests (#190) (@CarloLucibello)
• properly use or create buffer in DataLoader (#191) (@CarloLucibello)
• rework DataLoader (#192) (@CarloLucibello)
• make DataLoader iterations type stable (#193) (@CarloLucibello)
• relax joinobs to join different types (#194) (@CarloLucibello)
• add stratified options to splitobs (#195) (@CarloLucibello)
• add slidingwindow (#196) (@CarloLucibello)
• add batch_sequence (#197) (@CarloLucibello)

Closed issues:

• Scope of this package (#2)
• Status of MLDataPattern porting (#3)
• Adding a similar functionality like sklearn.make_regression, make_moons etc. (#15)
• add better documentation for batch (#55)
• mapobs doesn't work with named tuples (#70)
• CuArrays should not follow fast paths in obsview (#72)
• Padded batching (#78)
• The fallback result of batch over AbstractArray{<:Number} (#85)
• Archiving DataLoaders.jl (#90)
• Porting rescale! from MLDataUtils ? (#128)
• add trues_like and zeros_like (#137)
• v0.4.0 introduced error for DataLoader which collates images (#139)
• batchseq with vectors of vectors (#140)
• Error during callback: MLUtilsdid not define the expected moduleMLUtils`, (#152)
• mapobs(f, DataLoader) should just work? (#153)
• getobs!(buffer, data::BatchView, idx) ignores buffer (#156)
• joinobs assumes all inputs to be of the same type but returns a different JoinedData type (#160)
• Un-documented behaviour of splitobs(...; at=1) (#166)
• splitobs make views, but say it requires only getobs (#167)
• Type instability in DataLoader (#168)
• Floops depedency should be removed (#175)
• The JoinedData container seems to be missing a getindex() method for indices other than Number (#177)
• Allow passing RNG instance to splitobs() for shuffle control (#179)

MLUtils v0.4.5

Diff since v0.4.4

Merged pull requests:

• Add size(::DataLoader) (#174) (@mcabbott)
• update unstake docstring (#176) (@NeroBlackstone)
• More unthunking in ∇chunk (#180) (@mcabbott)

MLUtils v0.4.4

Diff since v0.4.3

Merged pull requests:

• Add RNG support to randobs (#162) (@EssamWisam)
• batchsize() documentation (#163) (@NeroBlackstone)
• Add firstindex(::AbstractDataContainer) (#164) (@Pangoraw)
• Improve examples for splitobs (#165) (@mcabbott)
• CompatHelper: add new compat entry for Statistics at version 1, (keep existing compat) (#169) (@github-actions[bot])
• fix rpad and put! ambiguities (#172) (@CarloLucibello)
• fix print test (#173) (@CarloLucibello)

Closed issues:

• eachobsparallel CUDA Error while freeing... (#161)
• The chunk function is not differentiable on GPU (#170)
• Method ambiguities reported by Aqua (#171)

MLUtils v0.4.3

Diff since v0.4.2

Closed issues:

• TaskPoolEx leads to unreliable Dataloaders (#142)
• "failed to start primary task" with Julia 1.9 and nthreads(:interactive) > 0 (#158)

Merged pull requests:

• Change default parallel executor to ThreadedEx (#143) (@RomeoV)
• CompatHelper: bump compat for NNlib to 0.9, (keep existing compat) (#159) (@github-actions[bot])

MLUtils v0.4.2

Diff since v0.4.1

Closed issues:

• unstack not passing the inferred test (#149)
• Using DataLoader for multi input (#151)

Merged pull requests:

• add a type-stable unstack method with Val dims arg (#150) (@gabrevaya)
• CompatHelper: bump compat for StatsBase to 0.34, (keep existing compat) (#154) (@github-actions[bot])

MLUtils v0.4.1

Diff since v0.4.0

Merged pull requests:

• update workflows (#146) (@CarloLucibello)
• define mapobs behavior for vector of indexes (#147) (@CarloLucibello)

MLUtils v0.4.0

Diff since v0.3.1

Closed issues:

• broken inferred tests for stack and batch (#52)
• chunk returns views for CuArrays (#103)
• Extend chunk to take a Vector of sizes (#114)
• Accomodate Base.stack (#119)
• stack can cause StackOverflowError or Segmentation fault on macOS ARM (#121)

Merged pull requests:

• improvements to stack (#125) (@CarloLucibello)
• improvements to chunk (#133) (@CarloLucibello)
• chunk by partition indexes (#134) (@CarloLucibello)

MLUtils v0.3.1

Diff since v0.3.0

Closed issues:

• out of date compatibility (#129)
• MLUtils exports rpad (#130)

Merged pull requests:

• deprecate rpad correctly (#131) (@CarloLucibello)