by

Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Dan Dixey
    @dandxy89
    Hello Nejc, this is a project I'd like to start help contributing on. I've raised an idea on the main repo to use a 'better' benchmarking library
    So I can start looking into that if you'd like?
    Would you also be open to considering the use compute.scala as an additional backend? https://github.com/ThoughtWorksInc/Compute.scala
    Nejc Ilenic
    @inejc
    Hey Dan, sorry for a late reply, needed to enable the notifications. Yeah, better benchmarking would be really useful, the current solution is pretty naive.
    Nejc Ilenic
    @inejc
    Regarding an additional backend: not sure what would the benefit be at this stage. For an MVP I think it might be more useful to focus on implementing the features we know are much needed in an ML setting (i.e. feature preprocessing, ensemble methods, unsupervised, ...). What do you think? I'm open to suggestions.
    Rok Novosel
    @novoselrok
    I would be up for implementing some the clustering methods, k-means, dbscan, hierarchical clustering
    I would also like to try and implement the parallel versions of these algorithms
    Rok Novosel
    @novoselrok
    (to try and beat scikit in speed at least :smile: )
    Nejc Ilenic
    @inejc
    Hey Rok, sounds great! Let's create a new package called clustering?
    Rok Novosel
    @novoselrok
    Yep, sounds good
    Nejc Ilenic
    @inejc
    Rok Novosel
    @novoselrok
    awesome! :thumbsup:
    Dan Dixey
    @dandxy89
    The website looks really good! I'm going to get back on with the benchmarking (moving houses atm)
    Nejc Ilenic
    @inejc
    Sounds great, no rush :smile:
    Matej Klemen
    @matejklemen
    Hey, quick question: is there going to be some more custom variable types created along the line? What I mean by that is something like "Discrete", "Continuous" or something like that, which would group multiple types like Int, String, Char, etc.. Since (for example) classification metrics are not always associated with doubles.
    Or will the user be expected to transform everything into doubles?
    Nejc Ilenic
    @inejc
    Hey @matejklemen, sorry for a late reply. Currently, the project is more or less halted due to lack of time.
    Matej Klemen
    @matejklemen
    @inejc ah that's unfortunate. Please let me know when/if you decide to start actively developing it again
    Nejc Ilenic
    @inejc
    @matejklemen I'm planning to work on it again, but this is a side project and thus a low priority one. In any case, would love to hear how you are/are plannning to use it.
    Matej Klemen
    @matejklemen
    @inejc Oh, I am just looking for a project that I can help develop whenever I have some spare time. Not actually using it for anything specific
    Nejc Ilenic
    @inejc
    In that case I'd be happy to offer any help you need with that
    Matej Klemen
    @matejklemen
    Alright :) I'll submit a PR when I have something ready.
    I'm still curious about my original question - are the Doubles as types for features and targets (e.g. type Features = DenseMatrix[Double]) just "placeholders" (to have some kind of foundation, and will be expanded later) or will everything need to be transformed into Doubles before working with the library (kind of like sklearn mostly just assumes data is numeric)
    Nejc Ilenic
    @inejc
    There are currently no plans to support other data types, although that is subotimal. For the next steps I wanted to focus on the API itself again (rewrite in a more functional manner), look at this PR for example: picnicml/doddle-model#32 and to write wrappers for lib-linear and lib-svm
    matrixbot
    @matrixbot
    @toxicafunk:matrix.org hi
    toxicafunk
    @toxicafunk
    has this project moved or is it just... dead?
    Matej Klemen
    @matejklemen

    Hey @toxicafunk, at the moment both me and @inejc are quite busy with other stuff, so the project is not getting any attention.

    Were you planning on using this library for some project of yours?

    toxicafunk
    @toxicafunk
    we are starting to get into the whole ai/deeplearning for search thing, we are using scala in part of our codebase (including spark) and was thinking this could be a good project to get started with ML