Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Zheng Xu
    @XericZephyr
    The function of shuffle in ImageRecordIter is currently weak.
    So you can do shuffle the lines in the image list several times.
    Sheng Wang
    @shengwa
    Data shuffling may be one problem. You can also test different learning rates.
    Kornél
    @korneil
    data is shuffled. what is the range of learning rate you suggest to try out?
    sunsocool
    @sunsocool
    Hello Guys. I am founding a Deep Learning International WeChat group. I would like to invite you to join this WeChat group.
    The mission of this group is not calling for papers but calling for awesome products.
    I wrote the group member introduction in this post and the link is:
    https://www.linkedin.com/pulse/deep-learning-international-wechat-group-welcome-join-lin-sun
    My WeChat ID is sunsocool, Add my WeChat with message "MxNet" and I'll send you the group invitation.
    Thank you guys. Welcome to join the group!
    Qingsong Liu
    @pineking
    @sunsocool
    Tong Guo
    @guotong1988
    @sunsocool
    Deep Learning International WeChat Group,I add
    sunsocool
    @sunsocool
    @guotong1988 @pineking Welcome you guys join Deep Learning International WeChat group~ :clap:
    Parag K Mital
    @pkmital
    Hi I'm interested in using the pretrained 21k ImageNet model, though not sure how to get started. I have installed mxnet, and found the model here: https://github.com/dmlc/mxnet-model-gallery/tree/master/imagenet-21k-inception - though not sure how to load the parameters into a network for prediction?
    Parag K Mital
    @pkmital
    ok, nevermind, figured it out... first create a FeedForward net, then call load with the string 'Inception' and the epoch number...
    sunsocool
    @sunsocool
    @tw991 Welcome you join Deep Learning International WeChat group~ :clap:
    Tian Wang
    @tw991
    @sunsocool Deep Learning International WeChat Group. Thanks for inviting!
    Kornél
    @korneil
    i wrote a new operator for test purposes. how can I include in the build?
    oh I see Makefile reads it
    Chris DuBois
    @doobwa
    What happened to mx.sym.Embedding? It's still used by https://github.com/dmlc/mxnet/blob/master/example/rnn/lstm.py, but I don't see it in v0.5.0.
    Zheng Xu
    @XericZephyr
    I think you need to git pull the newest mxnet source code and rebuild the entire project. Embedding seems to be a new operator in C++ implementation.
    sunsocool
    @sunsocool
    @XericZephyr Welcome you join Deep Learning International WeChat group~ :clap:
    baoqp
    @baoqingping
    @sunsocool Deep Learning International WeChat Group. Thanks for inviting!
    sunsocool
    @sunsocool
    @baoqingping Welcome you join Deep Learning International WeChat group~ :clap:
    Chris DuBois
    @doobwa
    @XericZephyr That worked. Thanks!
    Another question: I'm trying to use my compiled MXNet from Julia. I set MXNET_HOME, but I get:
    INFO: MXNET_HOME environment detected: /home/chris/mxnet INFO: Trying to load existing libmxnet... INFO: Failed to load existing libmxnet, trying to build from source...
    Any ideas why a given libmxnet.so might fail to load?
    baoqp
    @baoqingping
    Does anyone run the example adversary. I met an error
    baoqp
    @baoqingping

    for data, label in train_iter:
    ValueError: too many values to unpack
    then I modified the code as follows:

    for data, label, pad ,index in train_iter:

    another error occured: TypeError: type <type 'list'> not supported at arg_map["data"][:] = data
    again I modified the code and it worked

    for data, label, pad ,index in train_iter:
            data = train_iter.getdata()
            label = train_iter.getlabel()

    I wonder is it necessory to use getdata() and getlabel()

    Tianqi Chen
    @tqchen
    @doobwa This means you did not build mxnet in the path
    you can open an issue on MXNet.jl to ask chiyuan
    Chris DuBois
    @doobwa
    @tqchen I'm pretty certain I built mxnet, and exported MXNET_HOME. Started an issue here: dmlc/MXNet.jl#41. Thanks!
    zhubuntu
    @zhubuntu
    This message was deleted
    This message was deleted
    This message was deleted
    This message was deleted
    zhubuntu
    @zhubuntu

    i can't enter the main function with this code in my program

    mshadow::TShape f;
        f = mshadow::Shape2(3, 5);
        mxnet::Context d = mxnet::Context::Create(mxnet::Context::kCPU, 1);
        mxnet::NDArray a(f, d, false);

    but it worked with

    mxnet::NDArray c;

    who can help me

    sunsocool
    @sunsocool
    @zhubuntu How did you delete your chat history. Why I can't delete my Chat history?
    zhubuntu
    @zhubuntu
    @sunsocool put the mouse on the right top of the message you send.
    sunsocool
    @sunsocool
    @zhubuntu My Edit and Delete button both grey , I cannot click the button~
    sunsocool
    @sunsocool
    This message was deleted
    sunsocool
    @sunsocool
    @zhubuntu I seems that I can only delete the lastest message I just sent. I cannot delete it when it is not the lastest message in the room.
    Hang Su
    @suhangpro

    I found two examples about making predictions using pre-trained models:
    1) mxnet/example/notebooks/predict-with-pretrained-model.ipynb
    2) mxnet/tests/python/predict/mxnet_predict_example.py

    In 2) there is an explicit conversion from rgb to bgr:

    sample = sample[:, :, [2,1,0]]
    however there's no such conversion in 1). Which one should be fixed?

    freepjf
    @freepjf
    i want to know can there any w
    can there any c/c++ exmples for training with mxnet
    i can see only phython and java exmples in the doc but no c,
    Brian C
    @NOTtheMessiah
    don't have enough RAM to run deepstyle above max-edge-width 350, what can other consumer video cards run deepstyle at?
    maybe a better question might be how does RAM usage scale with width? I assume O(X^2) for width X, which should be easy to test
    Brian C
    @NOTtheMessiah
    Why does out.infer_shape(data=(1, 3, 192, 192)) work within Python, but mx.infer_shape(out,data=(1,3, 192,192)) doesn't seem to work in Julia?

    this is what I'm getting:
    ./dmlc-core/include/dmlc/logging.h:208: [14:35:29] src/operator/./convolution-inl.h:323: Check failed: ksize_x <= dshape[3] && ksize_y <= dshape[2] kernel size exceed input
    LoadError: MXNet.mx.MXError("[14:35:29] src/operator/./convolution-inl.h:323: Check failed: ksize_x <= dshape[3] && ksize_y <= dshape[2] kernel size exceed input")
    while loading In[7], in expression starting on line 1

    [inlined code] from ~/.julia/v0.4/MXNet/src/base.jl:57
    in infer_shape#3 at ~/.julia/v0.4/MXNet/src/symbolic-node.jl:253

    Brian C
    @NOTtheMessiah
    I figured it out, it seems to be backwards, because mx.infer_shape(out,data=(192,192,1,3)) gets no errors
    Adam
    @AdamHibble
    Hey guys,
    What is the status on the scala team? I am writing a deep learning with Scala series, would love to include mxnet.
    Yuan (Terry) Tang
    @terrytangyuan
    Hi Adam, thanks for your interest! You can watch our progress in my forked copy of mxnet.
    We'll push it to mxnet main repo once it's ready