v0.4.3

Yota v0.4.3

Diff since v0.4.2

Merged pull requests:

• Scatter ops (#74) (@dfdx)

v0.4.2

Yota v0.4.2

Diff since v0.4.1

v0.4.1

Yota v0.4.1

Diff since v0.4.0

Yota v0.4.1

• fix type inference issue on Julia 1.5

v0.4.0

Yota v0.4.0

Diff since v0.3.0

Yota v0.4.0

• new default tracer based on IRTools
• migration from CuArrays.jl to CUDA.jl
• new minimal Julia version is now 1.4

Closed issues:

• cannot differentiate getindex(::Tuple, ::Int) (#61)
• Help with implementing diffrule for element-wise division (#68)

Merged pull requests:

• Tuple getindex diffrule (#62) (@dfdx)
• Irtools tracer (#67) (@dfdx)
• fix typo in docstring (#71) (@oxinabox)

v0.3.0

Yota v0.3.0

Diff since v0.2.4

Derivatives w.r.t. constructor arguments, compatibility with Julia 1.4 and more.

Closed issues:

• Not able to build objects inside functions (#55)
• Performance question (#57)
• Errors when broadcasting (#58)

Merged pull requests:

• Raise bounds on Cassette & CuArrays (#59) (@mcabbott)
• Derivatives w.r.t. constructor arguments (#60) (@dfdx)

v0.2.4

Yota v0.2.4

Diff since v0.2.3

Merged pull requests:

• Install TagBot as a GitHub Action (#53) (@JuliaTagBot)
• Fix to device int (#54) (@dfdx)

v0.2.3

v0.2.3 (2019-12-28)

Diff since v0.2.2

Merged pull requests:

• Run eliminate_common() after backward pass (#52) (dfdx)
• Updated versions (#51) (dfdx)

v0.2.2

Better support for loops and array slicing/concatenation

v0.2.1

Update device API

v0.2.0

v0.2.0 (2019-09-21)

Diff since v0.1.3

Closed issues:

• Trying to differentiate imfilter convolution (#39)
• Can't use Keyword Arguments (#37)
• Explore possible solutions for dynamic graph support (#29)
• Add a note about broadcasting and primitives to README (#25)
• Add derivative for diag and vec (#24)
• Binarize *, /, + and - (#23)
• Type instability in Tape? (#21)
• User-level way to define primitive functions (#2)

Merged pull requests:

• Include version and compat section to Project.toml (#45) (dfdx)
• Set version in Project.toml (#44) (dfdx)
• Make nodiff work with broadcasting (#43) (dfdx)
• Back to Cassette + some refactoring (#42) (dfdx)
• Add tracing support for callable objects as functions (#41) (dfdx)
• New tracer based on JuliaInterpreter (#40) (dfdx)
• Make it possible to use unpacked tuples in trace() & grad() (#38) (dfdx)
• Tracing and backpropagation through %new nodes (#36) (dfdx)
• simplegrad: gradient function with tape, buffers and cached stripped out (#33) (dfdx)
• Dynamic grad (#32) (dfdx)
• Fixed type stability (#31) (dfdx)
• Diag vec rules (#30) (dfdx)
• Binarize *, /, etc. (#27) (dfdx)