Hi Mike, Thanks for the update.
> Opening the source is not entirely my decision, this is a collaboration > with the Thinktopic folks (Jeff Rose et al.). I'm personally in favour of > being pretty open about this stuff but I do think that it would be a > mistake if people build too much stuff on top of the wrong API which is > likely to happen if we release prematurely. > > I believe that number of users that would try this is small anyway, and they will use exclusively for evaluation and hobby at first, so, as Kovas said, having people build too much on it is something we can only hope for! :) I hope the Thinktopic folks would support opening it... > Things I'm paricularly keen to have nailed down in particular before we go > public: > 1. A tensorflow-like DAG model that allows arbitrary operations to be > composed compose into (possibly nested) graphs > 2. Enough abstraction that different backends work (ClojureScript, > Pure-JVM, Native, GPU etc.). core.matrix provides most of this, but there > are still some deep-learning specific operations that need tuning and can > be quite backend-specific, e.g. cuDNN has some specific ways of dealing > with mini-batches which we need to get right. I'd love to try Neanderthal > with this too if we can get the core.matrix integration working. > 3. Decent, stable APIs for the algorithms that you typically want to run > for mini-batch training, cross-validation etc. > 4. Pluggable gradient optimisation methods (we currently have stuff like > ADADELTA, SDG, ADAM etc. but would like to make sure this is sufficiently > general to support any optimisation method) > > I'll have a talk with the Thinktopic folks and see if we can come up with > a timeline for a open source release. In the meantime, if anyone is > *really* interested then we may be able to arrange collaboration on a > private basis. > It's not urgent, so any rough estimate would be great! -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
