+1 to Dragan's inquiry. FWIW, was reviewing the state of affairs the other day:
- MXNet currently has the best JVM interop story, among DL frameworks that have competitive perf. - DL4J has improved a lot recently but still looks like it has a ways to go in terms of perf. Right now I'm more interesting in word2vec type things, which don't require a deep net, so I might give Neanderthal a shot. By the way, I'd love to see matrix/tensor benchmarks of Neanderthal and Vectorz vs ND4J, MXNet's NDArray, and BidMat.. :) On Thu, Oct 6, 2016 at 8:44 AM, Dragan Djuric <draga...@gmail.com> wrote: > Hey Mike, > > A friend asked me if I know of any good (usable) deep learning libraries > for Clojure. I remembered you had some earlier neural networks library that > was at least OK for experimenting, but seems abandoned for your current > work in a similar domain. A bit of digging lead me to this post. > > I understand that this library may not be completely ready yet, but I > wandered wheter now you were able to give a better estimation of where it > stands in comparison with other DL offerings, like what deeplearning4j guys > are doing, or even with the established non-Java libraries such as Theano, > Torch, Caffe, and TensorFlow. What is the chance of you releasing it even > if it is not 100% ready? > > I get the reluctance to commit to a certain API, but I don't think > everyone will rush to commit their code to the API you release anyway, and > the open development will certainly help both the (potential) users and > your team (by returning free testing & feedback). > > > On Tuesday, May 31, 2016 at 7:17:35 AM UTC+2, Mikera wrote: >> >> I've been working with a number of collaborators on a deep learning >> library for Clojure. >> >> Some key features: >> - An abstract API for key machine learning functionality >> - Ability to declare graphs / stacks of operations (somewhat analogous to >> tensorflow) >> - Support for multiple underlying implementations (ClojureScript, JVM, >> CPU, GPU) >> - Integration with core.matrix for N-dimensional data processing >> >> We intend to release as open source. We haven't released yet because we >> want to get the API right first but it is looking very promising. >> >> On Tuesday, 31 May 2016 02:34:41 UTC+8, kovasb wrote: >>> >>> Anyone seriously working on deep learning with Clojure? >>> >>> I'm working with Torch at the day job, and have done work integrating >>> Tensorflow into Clojure, so I'm fairly familiar with the challenges of what >>> needs to be done. A bit too much to bite off on my own in my spare time. >>> >>> So is anyone out there familiar enough with these tools to have a >>> sensible conversation of what could be done in Clojure? >>> >>> The main question on my mind is: what level of abstraction would be >>> useful? >>> >>> All the existing tools have several layers of abstraction. In >>> Tensorflow, at the bottom theres the DAG of operations, and above that a >>> high-level library of python constructs to build the DAG (and now of course >>> libraries going higher still). In Torch, its more complicated: there's the >>> excellent tensor library at the bottom; the NN modules that are widely >>> used; and various non-orthogonal libraries and modules stack on top of >>> those. >>> >>> One could try to integrate at the bottom layer, and then re-invent the >>> layers above that in Clojure. Or one could try to integrate at the higher >>> layers, which is more complicated, but gives more leverage from the >>> existing ecosystem. >>> >>> Any thoughts? >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> -- > 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. > -- 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.
