Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • 10:31
    orausch edited #8637
  • 10:31
    orausch opened #8637
  • 03:41
    saudet review_requested #8636
  • 00:09
    aboulang2002 opened #8636
  • Jan 23 22:07
    MarkusKramer commented #6409
  • Jan 23 11:40
    AlexDBlack opened #8635
  • Jan 23 11:03
    raver119 labeled #8634
  • Jan 23 11:02
    raver119 reopened #8634
  • Jan 23 11:02
    raver119 commented #8634
  • Jan 23 11:01
    orausch commented #8634
  • Jan 23 10:58
    raver119 closed #8634
  • Jan 23 10:58
    raver119 commented #8634
  • Jan 23 10:55
    orausch edited #8634
  • Jan 23 10:55
    orausch opened #8634
  • Jan 23 06:47
    raver119 labeled #8633
  • Jan 23 06:34
    raver119 commented #8633
  • Jan 23 06:31
    raver119 labeled #8633
  • Jan 23 06:31
    raver119 assigned #8633
  • Jan 23 06:04
    lpaatero commented #8633
  • Jan 22 19:40
    lpaatero commented #8633
raguenets
@montardon
@thofer I think the best chance to get an answer is to fill in an issue showing what you expect and what you have.
ritclifcoder
@ritclifcoder
Hello my friends I need to use k - means clustering where can I fidn examples
Zijing Zhang
@zzj0402_gitlab
Weka has some k means clustering algorithm
LiNKeR
@54LiNKeR
Please How can I configure ObjectDetectionReader with a MultiDataSetIterator? specifically how do I determine what indexes I pass in for the Iterator's addInput() and addOutput()?
ritclifcoder
@ritclifcoder
I c ant use weka in my android app
Christian Spriegel
@horschi
Hi, does beta6 not work with avx1 any more?
I am getting "libnd4j binary was built with AVX/AVX2 support, but current CPU doesn't have this instruction set."
but nd4j-native beta6 with classifier=linux-x86_64 is included
also it seems beta6 is preferring openblas over mkl. is that correct?
Patrick Burkart
@Maverick283_gitlab
Hi everyone, are there any good guides on how to create a custom dataset iterator for regression? I got some but I do struggle to override "totalOutcomes" as for regression there is no fixed number. What should be put there?
aradford
@aradford_twitter
I'm getting an error in the Maven build. Access denied to http://maven.restlet.org/org/webjars/bower/graphlib/maven-metadata.xml. Error code 403, Forbidden. Any ideas?
Richard Eckart de Castilho
@reckart
hi there
could it be that DL4J beta 6 is not smart enough to fall back to a generic CPU binaries if both the AVX2 and generic binaries are on the classpath and the CPU does not support AVX?
Richard Eckart de Castilho
@reckart
chris-bamford
@chris-bamford
Hi there. I am new to ND4J and am trying to get to grips with the CSR Sparse Matrix (createSparseCSR()). I have created one as an INDArray and would like to print it out to ensure I have done it correctly. However, the normal toString() just prints
org.nd4j.linalg.cpu.nativecpu.SparseNDArrayCSR@59546cfe
Also getRow() and getDouble() return only zeroes. Any tips?
chris-bamford
@chris-bamford
(The deeplearning4j-examples project sadly does not cover createSparseCSR or createSparseCOO)
chris-bamford
@chris-bamford
Ah - I have answered my own question - you need to call matrix.toDense() first to return a regular INDArray on which you can perform normal operations.
Martin Krybus
@dobretotamje_gitlab
Hi everyone, just updated my pom to beta6 in my ALE project and now getting exception on wrong input (using DQNFactory): Got rank 3 array as input to ConvolutionLayer (layer name = layer0, layer index = 0) with shape [1, 120, 40]. Expected rank 4 array with shape [minibatchSize, layerInputDepth, inputHeight, inputWidth]. (layer name: layer0, layer index: 0, layer type: ConvolutionLayer), can someone direct me please?
Samuel Audet
@saudet
@dobretotamje_gitlab please post an !issue about that with a short code snippet to reproduce that error
f
@SidneyLann
Is sd.nn.batchNorm(...) ready to be used?
Bjørn Magnus Mathisen
@epichub
So I am trying to patch the KerasImport code to support multiple inputs per layer, would the easiest way be to insert a StackingVertex before a layer with two inputs (and a corresponding unstackingvertex if needed)?
padriba
@padriba
Hi
I have built a keras CNN model, the problem is despite the model kept converging but it becomes too slow 0.0042 by epoch since the loss = 4.
how can i speed up my model convergence ?
i will provide you the code if needed
daanvdn
@daanvdn
hi could somebody tell me if org.deeplearning4j.models.node2vec.Node2Vec is usable? It's been marked as 'under construction' for quite a few releases. Does anybody have an idea if a functional dl4j implementation of Node2vec is underway?
Dominik
@domino2

I'm building the network to validate the directional graph structure.
So I have matrices to reflect the structure and edge properties.
On the end I'm expecting to have a network works as a heuristic function to roughly evaluate the graph to say which one is "better".

I wanted to have all information in one file line, read it by CSV Reader, create a schema and pull to network (train, experiment with and so on).

I can have in one line more then 1000 cells which need to proceed be the network after all.

Can somebody validate my thoughts or give some feedback?
Maybe somebody did similar task or has better overview how to model this.

Thank you

Bjørn Magnus Mathisen
@epichub
Any devs here?
I may have a patch for modelimporter which makes it possible to import Siamese networks from (tf.)Keras
It probably needs som work but it does work for importing my Siamese network with a lambda layer for combining the two embeddings (basically my keras model is implementing Figure 5 from this paper https://link.springer.com/article/10.1007/s13748-019-00201-2 )
Matt Davis
@mdavis95
I would be interested in seeing the
Patch. Just implemented a Siamese network using stack and unstack and would like to see alternative approaches
Bjørn Magnus Mathisen
@epichub
that is what the patch does
detects when two input signal is being forwarded through the same layers in the keras config
then inserts stack and unstack layers in between at the appropriate places while reading the keras config
Christian Malek
@christianmalek
The links to https://docs.skymind.ai/docs in the README.md are stale
Bjørn Magnus Mathisen
@epichub
however I had to create my own layer and mapping in getKerasLayerFromConfig in KerasLayerUtils to be able to get my lambda to accept two inputs
that is a bad hack
Christian Malek
@christianmalek
The link to the twitter acc https://twitter.com/deeplearning4j is stale, too. Your account is suspended
Bjørn Magnus Mathisen
@epichub
ideally I would like to extend SameDiff to accept two inputs, so that it can be added dynamically as a mapping as "normal" lambdas are
Christian Malek
@christianmalek
I've just got a message from a bot. So Gitter is dead and this konduit page is the new place to ask/help with questions regarding DL4J?
Bjørn Magnus Mathisen
@epichub
afaik yes
there still seems to be quite a number of people lurking here though
Christian Malek
@christianmalek
Thanks @epichub
Bjørn Magnus Mathisen
@epichub
Foodyborris
@Foodyborris
Hi. I am wanting to vary the L2 regularization factor part way through training. I don't see an easy way of doing this (say versus changing the learning rate). How would you recommend I approach this? Do I need to make a suitable NeuralNetConfiguration and attach it to the layers of the existing model? MTIA.
Birdasaur
@Birdasaur
Hello. We need to obtain the Jacobian Matrix with respect to the input of a MultiLayerNetwork model to the outputs. (not with respect to the weights) Is there a existing method or approach that would achieve this using DL4J? Has anyone dones this before?
Our goal is to mimic the example shown here: https://medium.com/unit8-machine-learning-publication/computing-the-jacobian-matrix-of-a-neural-network-in-python-4f162e5db180 which demonstrates using Tensorflow or Autograd.
Roy Meshulam, M.Sc.
@roymeshulam_twitter
Anyone used dl4j for fx trading? Interested to get your experience