chad-fast - [no description]

	Commit message (Collapse)	Author	Age
*	Benchmark GMM	Tom Smeding	2024-11-14
\|
*	Remove warnings	Tom Smeding	2024-11-13
\|
*	Make EBuild derivative aware of zero cotangent arrays	Tom Smeding	2024-11-11
\|
*	Test gmm	Tom Smeding	2024-11-10
\|
*	Fix forward derivative of 'exp'	Tom Smeding	2024-11-10
\|
*	Test GMM; it fails	Tom Smeding	2024-11-10
\|
*	Fix bug in simplifier (effects are hard)	Tom Smeding	2024-11-10
\|
*	Complete GMM implementation	Tom Smeding	2024-11-10
\|
*	Cleanup, more Language operations	Tom Smeding	2024-11-09
\|
*	Add Custom to Language	Tom Smeding	2024-11-09
\|
*	Some more primitive operators	Tom Smeding	2024-11-09
\|
*	Maximum/minimum	Tom Smeding	2024-11-09
\|
*	WIP maximum/minimum	Tom Smeding	2024-11-08
\|
*	Custom derivatives	Tom Smeding	2024-11-08
\|
*	WIP custom derivatives	Tom Smeding	2024-11-08
\|
*	Remove build1	Tom Smeding	2024-11-07
\|
*	Benchmark	Tom Smeding	2024-11-07
\|
*	Some simplification rules	Tom Smeding	2024-11-06
\|
*	Generate EOneHot in D[EIdx]	Tom Smeding	2024-11-05
\| \| \| \| \| \|	This generates a one-hot for the zero-dimensional inner array because indexing one level further to the actual element is too difficult. But this should simplify away fine.
*	Support EOneHot	Tom Smeding	2024-11-05
\|
*	WIP EOneHot	Tom Smeding	2024-11-04
\|
*	simplifyFix	Tom Smeding	2024-10-29
\|
*	Store only on tape what's used in the dual	Tom Smeding	2024-10-29
\|
*	WIP preserve only subset of D0 bindings in dual (...)	Tom Smeding	2024-10-27
\| \| \| \| \| \| \| \| \| \| \| \| \| \| \|	The point of this is to ensure that when an expression occurs in a Build, then the parts of D0 that are only there to make sharing work out for D1 are not laboriously taped in an array and preserved for D2, only for D2 to ignore them. However, while the subtape machinery is a good first step, this is not everything: the current Build translation makes a Build for the (elementwise) tape and separately a build for the primal. Because the primal _does_ generally need the subtaped-away stuff, we can't just not tape those. TODO: figure out how to resolve this / what the next step is.
*	Fix build	Tom Smeding	2024-10-27
\|
*	Better inline syntax for Language	Tom Smeding	2024-10-27
\|
*	Fix EIdx derivative	Tom Smeding	2024-10-26
\|
*	Show accumulator as dummy string, not error	Tom Smeding	2024-10-26
\|
*	Fix IIxNil warning	Tom Smeding	2024-10-26
\|
*	Make interpreter debug printing conditional	Tom Smeding	2024-10-26
\|
*	Fix {} usage in pretty-printing of ELet	Tom Smeding	2024-10-26
\|
*	Debugging	Tom Smeding	2024-10-26
\|
*	Fix interpreter bug	Tom Smeding	2024-10-22
\|
*	Less warnings	Tom Smeding	2024-10-22
\|
*	Differentiate Replicate	Tom Smeding	2024-10-22
\|
*	Tests	Tom Smeding	2024-10-21
\|
*	WIP testing neural	Tom Smeding	2024-10-18
\|
*	More towards test suite	Tom Smeding	2024-10-14
\|
*	Towards a test suite	Tom Smeding	2024-10-07
\|
*	Reverse-by-forward, and checking neural (it's wrong)	Tom Smeding	2024-10-01
\|
*	Dual-numbers forward AD	Tom Smeding	2024-09-28
\|
*	The size of the empty shape is 1, not 0	Tom Smeding	2024-09-26
\|
*	Test neural	Tom Smeding	2024-09-25
\|
*	Add some missing cases	Tom Smeding	2024-09-22
\|
*	Interpreter typechecks, at the cost of compositionality of RepAc	Tom Smeding	2024-09-22
\|
*	Stuff	Tom Smeding	2024-09-22
\|
*	Some hacking	Tom Smeding	2024-09-22
\|
*	A little interpreter accum progress	Tom Smeding	2024-09-16
\|
*	WIP Accum stuff	Tom Smeding	2024-09-15
\|
*	WIP better zero/plus, fixing Accum (...)	Tom Smeding	2024-09-13
\| \| \| \| \| \| \| \| \| \| \| \| \|	The accumulator implementation was wrong because it forgot (in accumAdd) to take into account that values may be variably-sized. Furthermore, it was also complexity-inefficient because it did not build up a sparse value. Thus let's go for the Haskell-interpreter-equivalent of what a real, fast, compiled implementation would do: just a tree with mutable variables. In practice one can decide to indeed flatten parts of that tree, i.e. using a tree representation for nested pairs is bad, but that should have been done _before_ execution and for _all_ occurrences of that type fragment, not live at runtime by the accumulator implementation.