Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Daniel Ryan
    @dan-ryan
    Still can't run my code. I've created a bug: liquidcarrot/carrot#115
    Unless I'm missing something and there is another way?
    Daniel Ryan
    @dan-ryan
    What would be the best way to build a NN to check if a graph is cyclic or a forest. I've been thinking how to input a graph of different sizes and shapes and unsure how to best do this.
    Daniel Ryan
    @dan-ryan
    For those who never did graph theory:
    https://en.wikipedia.org/wiki/Cyclic_graph
    Daniel Ryan
    @dan-ryan
    Haven't thought up a good solution. I've put up my question here: liquidcarrot/carrot#135
    Luis Carbonell
    @luiscarbonell

    For those who never did graph theory:
    https://en.wikipedia.org/wiki/Cyclic_graph

    Going to add (explicitly) to the Wiki to the issue for reference - thought it was good. :thumbsup:

    Christian Echevarria
    @christianechevarria
    Garfield100
    @Garfield100
    Hello, noobiest of noobs here, I was trying to do some image stuff with carrot but needed 4096 inputs and outputs (and more in between). Node ran out of memory, is it simply not yet possible to have so many nodes or is there something other than a perceptron i should be using? I can give more details if necessary, I mostly want to know if it's at all possible.
    (Luis, I got sick of the car thing, there are problems with my line intersection stuff I haven't yet managed to solve, and also I didn't feel like it would be very rewarding even if it ended up working, it's going on the long long list of unfinished projects for now)
    Christian Echevarria
    @christianechevarria
    Hey @Garfield100 Node has a default memory limit of 512mb
    You can set a flag to increase it
    This article should still work:
    Garfield100
    @Garfield100
    Ok, I guess doing image stuff simply takes forever, especially on a laptop thing. I've upped it to 2gb, let's see what happens
    Christian Echevarria
    @christianechevarria
    :pray:
    Feel free to reach out if you need any more help
    Garfield100
    @Garfield100
    Ran out of memory again :P thanks, but I'm afraid this needs to be done on something with a beefy GPU and more than a bit of memory of which a good part is being eaten by chrome
    Thanks for the help :)
    Christian Echevarria
    @christianechevarria

    Absolutely, NVIDIA has a tiny GPU rig for $99 that can be good for development:

    https://www.amazon.com/dp/B07SGBHDCZ/ref=cm_sw_em_r_mt_dp_U_3nkrDbM800KNF

    Garfield100
    @Garfield100
    That's super cool! Does carrot support using the GPU yet?
    Christian Echevarria
    @christianechevarria

    That's super cool! Does carrot support using the GPU yet?

    It's coming, we're doing a re-architecture of the library to clean up bugs and speed up development but that's top priority right after

    Garfield100
    @Garfield100
    Oo, awesome. I assume it will only be for NVIDIA cards again?
    Christian Echevarria
    @christianechevarria

    Oo, awesome. I assume it will only be for NVIDIA cards again?

    Not sure about this one, we were planning to build on top of GPU.js to implement the functionality but I know they're still in earlier stages of development

    Oo, awesome. I assume it will only be for NVIDIA cards again?

    One of our stargazer, Robert: https://github.com/robertleeplummerjr is a core contributor to GPU.js: https://github.com/gpujs/gpu.js -- seems like a good question to bring up

    Garfield100
    @Garfield100
    They have a nice online benchmark tester, so I did it using only the integrated graphics I have
    image.png
    This was the result
    So I'm guessing if even integrated graphics speeds it up, it should support any gpu. I couldn't find any mention of Nvidia or CUDA anywhere.
    Christian Echevarria
    @christianechevarria

    image.png

    Haha awesome!

    So I'm guessing if even integrated graphics speeds it up, it should support any gpu. I couldn't find any mention of Nvidia or CUDA anywhere.

    I looked too and I couldn't find anything that scoped support

    Christian Echevarria
    @christianechevarria

    Hi everyone! :wave:

    Some updates:

    We're continuing work on a complete refactor of the library with all of the suggestions and issues y'all added in the main repo --

    For now, we're using this repository as our sandbox / playground:
    https://github.com/liquidcarrot/nn

    Happy to announce that we're approaching feature parity with the core repo; if you go to the sandbox repository, you'll see a rough timeline of features and milestones. During the refactor we'll be doing very minor maintenance changes to the main repository. Hopefully we'll be pushing the big update in about ~4 weeks time.

    In addition to keeping the same API (with some minor tweaks and enhancements) the new update will also support streaming :star:

    And we're developing a series of articles, walk-throughs, and idea boards in the repo's wiki:
    https://github.com/liquidcarrot/nn/wiki/Creating-Liquid-Carrot

    We'd love to hear your thoughts / get your input on all of it!

    P.S. One of our first priorities as soon as we get to feature parity is going to be GPU support, so I'll keep you guys updated on that :rocket:

    therealjaredcodling
    @therealjaredcodling
    Awesome stuff guys!
    Christian Echevarria
    @christianechevarria
    @therealjaredcodling thank you man! We have @luiscarbonell in a cave coding, but some good stuff is coming
    Luis Carbonell
    @luiscarbonell

    We have @luiscarbonell in a cave coding, but some good stuff is coming

    Hey guys just stepping out of the cave briefly to mention: we started a series/tutorial on how to create a neural network library from scratch in JavaScript - basically outlining and open-sourcing not only our code but our ideas as well. We just got this started, but here's part two of the series: https://medium.com/liquid-carrot/creating-a-neural-network-from-scratch-in-javascript-part-1-2-1e65dbcb1445?source=friends_link&sk=f2949440d4439fc70a1bd1ea319f38e1

    If you guys have any feedback or comments - please feel to drop them.

    Akash Samlal
    @AkashSamlal
    Fantastic article, Luis! Thank you for both of you guys to share to the community on your open source library, makes learning a hella lot easier when we got people like you guys in the community! Keep it up!
    Christian Echevarria
    @christianechevarria

    Fantastic article, Luis! Thank you for both of you guys to share to the community on your open source library, makes learning a hella lot easier when we got people like you guys in the community! Keep it up!

    Thanks for the love @AkashSamlal!

    Makes learning a hella lot easier when we got people like you guys in the community! Keep it up!
    Absolutely will :fist:

    Loaiabdalslam
    @loaiabdalslam
    Hey all
    Christian Echevarria
    @christianechevarria
    Hi @loaiabdalslam :wave:
    Loaiabdalslam
    @loaiabdalslam
    Hey Guys i just want to ask something if i need to init a new network with a exist weights formated in array for example how can i achieve , is pretrained model is enabled here
    @christianechevarria
    if there is a method let me know about it and if there is not let's create transefer Learning Module inside our project , and make carrot support ONNX
    Christian Echevarria
    @christianechevarria
    @loaiabdalslam
    if there is a method let me know about it and if there is not let's create transefer Learning Module inside our project , and make carrot support ONNX
    Thanks for bringing this up 👌 This is a seriously good idea, I'm going to add it as one of issues now
    Hey Guys i just want to ask something if i need to init a new network with a exist weights formated in array for example how can i achieve , is pretrained model is enabled here
    We have a .fromJSON & toJSON system right now that can import the structure of models now, but it's not 100% yet: liquidcarrot/carrot#152
    Christian Echevarria
    @christianechevarria

    @loaiabdalslam Just slapped an issue here: liquidcarrot/carrot#159

    I think it's a great idea to do this and would help a lot of people, I want to make it a priority -- where do you think we can start on getting this done?

    Also sidenote for @/all I've been working on a demo to showcase the library & its features here:

    https://liquidcarrot.io/example.flappy-bird/

    I'd love for it to be a great introduction to machine learning and also be a tutorial on how to use the library -- has anyone seen some good examples that we can model off for this?

    Christian Echevarria
    @christianechevarria

    @christianechevarria
    if there is a method let me know about it and if there is not let's create transefer Learning Module inside our project , and make carrot support ONNX

    @loaiabdalslam By the way! Here's the link to the docs on this:
    https://liquidcarrot.io/carrot/Network.html#.fromJSON