Implemented hyperdual space for partials computation, and more

ChristopherRabotin commented 6 years ago

Implemented hyperdual space for partials computation: leads to significantly faster computation, and requires a single call to the function to be derived
Extended vector function partial matrices from simply the gradient computation: causes a breaking change where the nabla functions are renamed partials since nabla refers specifically to the gradient computation, which is square
Changed the partials function to return the state and the function derivative: a single call to the function to return the state and its derivative may be useful in applications where such computation is especially heavy
Added {Add,Mul,Sub,Div}-Assign from std::ops to allow for +=, *=, -= and /= operations on duals and their parametrized type, and associated tests

Closes #4

ChristopherRabotin commented 6 years ago

I think that this should make dual_num ready for a crates.io release for version 0.2.0.

novacrazy commented 6 years ago

Everything looks pretty good to me, although I'll take a deeper look after work. Those Allocator bounds do seem a little excessive...

ChristopherRabotin commented 6 years ago

Good point! I had added a bunch of them initially when things were not working well, but many can be removed now.

novacrazy commented 6 years ago

I'll merge this in a few minutes after some quick super minor cleanups.

By the way, if you'd rather use LaTeX from within doc comments, we can use this hack to do so.

novacrazy commented 6 years ago

Sorry for the delay, just published version 0.2.0

novacrazy / dual_num