"Diffrence between Feature selection and Feature Weighting"

IwanIwan MemberPosts:9Contributor I
edited June 2019 inHelp

Hi, I'm curios to know what is the big difference between Feature Selection and Feature Weighting ?

Perhaps it's just the same. Right now ,I'm doing some project with Optimize weight (PSO). Is it the same with feature selection ?

Can you guys, tell me the basic concepts about Optimize weight (PSO), coz i don't have clue about it....

Some good reading will be fine....

Thanks.

Iwan

Tagged:

Answers

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, MemberPosts:1,761Unicorn

    Feature selection is usually done with some sort of Performance Measure. You try different combinations of your features and measure how good or bad the performance will be. At the end of the iterations, the feature selection operator (there are a few), will return the best combination of features with the best performace. It will typically be a reduced set of features for your downstream processing.

    Feature weighting just uses the selected algorithm (i.e. Weight by SVM) to see how much influence each feature has in a classification problem. You can use a Select By Weights operator to get the "top k" of features for your downstream processing.

    RandyLeBlanc
  • IwanIwan MemberPosts:9Contributor I

    Hi Thomas,


    Could you tell me more about optimize weight (PSO) ?

    I know that the output is the feature weight. Is it (PSO) automatically select the best weight to the next step (perhaps classification procedure) , or we have to use "Select by weight" operator ......
    I really need basic concepts about optimize weight PSO, and maybe an example of it.

    For answering my question, big thanks to you.

    Iwan

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, MemberPosts:1,761Unicorn

    The Optimize Weights (PSO) operator uses a particle swarm optimization method to generate the feature weights but this type of method is use to reduce your attributes. It's typically a minimization technique. You can then export the selected features and their weights downstream in the process and then use the weights to automatically select the features from your scoring set. I do this quite a bit, see the attached example.




















    < portSpacing端口="source_training" spacing="0"/>
    < portSpacing端口="sink_model" spacing="0"/>
    < portSpacing端口="sink_through 1" spacing="0"/>





    <运营商激活= " true "类= com“性能”patibility="5.0.000" expanded="true" height="76" name="Performance (2)" width="90" x="179" y="30"/>




    < portSpacing端口="source_model" spacing="0"/>
    < portSpacing端口="source_test set" spacing="0"/>
    < portSpacing端口="source_through 1" spacing="0"/>
    < portSpacing端口="sink_averagable 1" spacing="0"/>
    < portSpacing端口="sink_averagable 2" spacing="0"/>

    A cross-validation evaluating a decision tree model.






    < portSpacing端口="source_example set" spacing="0"/>
    < portSpacing端口="source_input 1" spacing="0"/>
    < portSpacing端口="sink_performance" spacing="0"/>








    < portSpacing端口="source_training" spacing="0"/>
    < portSpacing端口="sink_model" spacing="0"/>
    < portSpacing端口="sink_through 1" spacing="0"/>





    <运营商激活= " true "类= com“性能”patibility="5.0.000" expanded="true" height="76" name="Performance" width="90" x="179" y="30"/>




    < portSpacing端口="source_model" spacing="0"/>
    < portSpacing端口="source_test set" spacing="0"/>
    < portSpacing端口="source_through 1" spacing="0"/>
    < portSpacing端口="sink_averagable 1" spacing="0"/>
    < portSpacing端口="sink_averagable 2" spacing="0"/>

    A cross-validation evaluating a decision tree model.

















    < portSpacing端口="source_input 1" spacing="0"/>
    < portSpacing端口="sink_result 1" spacing="0"/>
    < portSpacing端口="sink_result 2" spacing="0"/>




Sign InorRegisterto comment.