Skip to content

v1.0.4

Choose a tag to compare

View all tags
@github-actions github-actions released this 03 Nov 21:52
· 12 commits to main since this release
v1.0.4
36effdb

RealLabelNormalization v1.0.4

Diff since v1.0.3

Breaking changes

  • GPU-compatible functors update: converts columnwise/rowwise functors to use NTuple storage so that ColumnwiseScaleBack{T,F,N} and RowwiseScaleBack{T,F,N} are isbits and can be passed directly to CUDA kernels. Scalar functors (MinMaxScaleBack, ZScoreScaleBack, LogScaleBack) remain bitstype. Constructors accept Vector{F} and convert to NTuple for convenience.

Notes:

  • Tuples are bitstype; Vector is not — this fixes the “not a bitstype” GPU error.
  • For large numbers of rows/columns prefer using raw parameter arrays (min_vals, max_vals, offsets, etc.) in kernels to avoid large-tuple compilation overhead.
  • Avoid wrapping tuple functors in non-isbits wrappers; extract per-row/col functors or use parameter arrays when targeting CUDA.
Assets 2
Loading