DeepLearning4J Extension - (Early Release, Needs Feedback)

JEdward

Hi guys,

We have created an extension to integrate DeepLearning4J into RapidMiner! 
What we need, before release on marketplace, is a small army of bug finders & volunteers to become fixers of the extension. 

Download here & help us out. 
https://www.rapidminerchina.com/2016/02/rapidminer-china-announces-rapidminer-deeplearning4j-integration/
(Please note, this is still underdevelopment so use in production at your own risk)

MartinLiebig

I've tested it and i need to say:
Thank you for this! It is already great! (Even if i run into some null pointers..)

JEdward

Although it needs pointed at a text file at the moment, I recommend trying out Word2Vec, it's actually pretty interesting way to look at relationships between words.  The hierarchical cluster tree also provides a fun way to navigate around them too.

<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<process version="7.0.000">
  <context>
    <input/>
    <output/>
    <macros/>
  </context>
  <operator activated="true" class="process" compatibility="7.0.000" expanded="true" name="Process">
    <process expanded="true">
      <operator activated="true" class="dl4j_extension:word_2_vec" compatibility="1.0.000" expanded="true" height="82" name="Word2Vec" width="90" x="45" y="85">
        <parameter key="file_path" value="C:\Users\user\.RapidMiner\repositories\Local Repository\data\raw_sentences.txt"/>
        <enumeration key="stop_words"/>
      </operator>
      <operator activated="true" class="set_role" compatibility="7.0.000" expanded="true" height="82" name="Set Role" width="90" x="179" y="85">
        <parameter key="attribute_name" value="Word"/>
        <parameter key="target_role" value="id"/>
        <list key="set_additional_roles"/>
      </operator>
      <operator activated="true" class="multiply" compatibility="7.0.000" expanded="true" height="103" name="Multiply" width="90" x="313" y="85"/>
      <operator activated="true" class="agglomerative_clustering" compatibility="7.0.000" expanded="true" height="82" name="Clustering" width="90" x="447" y="85">
        <parameter key="mode" value="CompleteLink"/>
        <parameter key="measure_types" value="NumericalMeasures"/>
        <parameter key="numerical_measure" value="CosineSimilarity"/>
      </operator>
      <operator activated="true" class="data_to_similarity_data" compatibility="7.0.000" expanded="true" height="68" name="Data to Similarity Data" width="90" x="447" y="187">
        <parameter key="measure_types" value="NumericalMeasures"/>
        <parameter key="numerical_measure" value="CosineSimilarity"/>
      </operator>
      <connect from_op="Word2Vec" from_port="vector" to_op="Set Role" to_port="example set input"/>
      <connect from_op="Set Role" from_port="example set output" to_op="Multiply" to_port="input"/>
      <connect from_op="Multiply" from_port="output 1" to_op="Clustering" to_port="example set"/>
      <connect from_op="Multiply" from_port="output 2" to_op="Data to Similarity Data" to_port="example set"/>
      <connect from_op="Clustering" from_port="cluster model" to_port="result 1"/>
      <connect from_op="Clustering" from_port="example set" to_port="result 2"/>
      <connect from_op="Data to Similarity Data" from_port="similarity example set" to_port="result 3"/>
      <portSpacing port="source_input 1" spacing="0"/>
      <portSpacing port="sink_result 1" spacing="42"/>
      <portSpacing port="sink_result 2" spacing="21"/>
      <portSpacing port="sink_result 3" spacing="42"/>
      <portSpacing port="sink_result 4" spacing="0"/>
    </process>
  </operator>
</process>

 

NeuralMarket

I must try this. Thanks for your hard work!

MartinLiebig

Hey John,

quick question: How can i apply the model returned by of word2vec?

~Martin

JEdward

It turned out to be less than a quick question.  ::)

I created a process based on the rules here: http://deeplearning4j.org/word2vec (which is a really nice explanation on how it 'should' work) I created a sample process which applies the similarity operators, aggregate, clustering & PCA on the result to try to demonstrate it. 
Trust me, it was a thing of beauty using the aggregate, differencing, append & similarity operators in various combinations. 

Unfortunately, it doesn't work with as I expected on the complete works of Shakespeare so we're debugging the operator to see if something needs to be tweaked internally. 

kenedy

I really like what you have shared here.

kayman

Hi, this seems to be no longer available, any way to still get hold of this?

MartinLiebig

Hi,

 

it's still compilable from GitHub: https://github.com/LostSummer233/rapidminer-extension-dl4j-pack

 

~Martin

KPL

It's been a while since your last post. Is RapidMiner going forward with this extension or dropping it? I'm looking for a recurrent NN plug-in for RM and this seemed to be it. Any other recommendations for RNN & RM?