Comments on: Automatic Differentiation with Scala

By: jmount

jmount — Thu, 15 Jul 2010 00:13:50 +0000

Follow-up article and newer code: Gradients via Reverse Accumulation.

By: jmount

jmount — Thu, 01 Jul 2010 18:16:17 +0000

Updated source and jar (some bug fixes, some extensions and some stuff I hope to write about): http://www.win-vector.com/dfiles/ScalaDiff-0.2.jar , http://www.win-vector.com/dfiles/AutoDiff-0.2.sbp and http://www.win-vector.com/dfiles/AutoDiff-0.2.advert .

By: Ismael Juma

Ismael Juma — Thu, 17 Jun 2010 00:12:50 +0000

Hey,

The following:

for(k <- 0 to (npoint-1))

could also be done as:

for(k <- 0 until npoint)

As said elsewhere, things could be made more concise by using a more functional style (with a potential performance impact in some cases).

Best,
Ismael

By: jmount

jmount — Tue, 15 Jun 2010 21:32:02 +0000

@Scott Locklin One reason the source is pretty wordy is my use of for-loops to add numbers (instead of thinking more functionally). But Scala isn't as concise as the more functionally oriented languages.

By: Scott Locklin

Scott Locklin — Tue, 15 Jun 2010 20:31:47 +0000

Well, that’s the first real exposition of Scala I’ve yet seen. I hadn’t realized how similar it was to OCaML/F#. While these strongly/statically typed languages seem to be …. the future …. the whole language lawyer aspect of OCaML (and presumably Scala) was what kept me away from it in the first place. I’d rather sling code and deal with type plumbing later.

The source seems pretty wordy. I guess reimplementing dual numbers is going to be; hopefully that source gets used somewhere so nobody has to do it again. I’m going to remember the dual numbers AD trick for later.