Hi julia-users, I've started working on a weighted finite-state transducer package for Julia, inspired by OpenFst and the Pyfst wrapper. I've put it up on Github now: https://github.com/oadams/Fst.jl
It's very rudimentary and untested at the moment, but I thought I'd let the world know sooner rather than later. At the moment it has: - Wfst and Arc types and a couple associated functions. - Display functions to create a graphical representation (Haven't looked into where IJulia is at at the moment with graphical outputs but it would be good to use this in the spirit of IPython.Display for quick and easy WFST prototyping) - A composition function (epsilon transitions not supported - working on that in the 'epsilon' branch) - A topological sort function. It's partly an educational endeavour, but I need to make it reliable and efficient so it can serve as a Julia alternative to OpenFst for some real-world NLP research I'm working on. If anyone's interested I'd love to hear your thoughts and criticisms. In particular it'd be great to get feedback on my coding style and design decisions. I put some of my design questions and tentative answers in a document in the top level directory. Thanks, Oliver On 10 October 2014 13:09, Oliver Adams <[email protected]> wrote: > That sounds like a good idea. > > On 9 October 2014 15:24, Stefan Karpinski <[email protected]> wrote: > >> I wonder if it makes sense to combine the two at all. Initially, it's >> probably safe to make a separate package and then consider factoring out >> common types and behaviors after the fact. Looking forward to seeing this! >> >> On Thu, Oct 9, 2014 at 6:01 AM, Oliver Adams <[email protected]> >> wrote: >> >>> Based on a cursory read of the linked page, there doesn't seem to be a >>> connection of practical significance. >>> >>> I'm going to start coding up a little package. I did note the >>> FiniteStateMachine <https://github.com/tensorjack/FiniteStateMachine.jl> >>> package, which is somewhat related but doesn't have transducers or support >>> many operations. >>> >>> On Tuesday, 7 October 2014 13:57:11 UTC+1, Stefan Karpinski wrote: >>>> >>>> Not that I'm aware of. Possibly dumb question that I can't seem to find >>>> the answer to: what connection, if any, is there between these transducers >>>> and Clojure's new concept >>>> <http://blog.cognitect.com/blog/2014/8/6/transducers-are-coming> by >>>> the same name? >>>> >>>> On Tue, Oct 7, 2014 at 5:42 AM, Oliver Adams <[email protected]> >>>> wrote: >>>> >>>>> Hi there. I'm interested in doing some work with weighted finite-state >>>>> transducers in Julia. If there's something in the spirit of OpenFst or the >>>>> Python wrapper pyfst, it'd be great to know. If not then it might be fun, >>>>> educational and maybe even practical to implement some stuff myself. But >>>>> I'd rather not reinvent the wheel - has anyone been working on a package >>>>> that does this sort of stuff? >>>>> >>>>> Thanks. >>>>> >>>> >>>> >> >
