sequences

[Deleted User][Deleted User] Posts: 0 Learner III
edited February 2020 in Help
Hello
Look at this chart please, can RM analyze sequence of images like this? ( each square has an image and they have a sequence like matrix)


Thank you in advance
mbs

Best Answers

  • rfuentealbarfuentealba Moderator, RapidMiner Certified Analyst, Member, University Professor Posts: 568 Unicorn
    Solution Accepted
    Hello @mbs,

    I would definitely take a look at imagemagick to process the images (resize, cut, etc...), put these into folders and use some more magic to transform the image into something texty.

    ImageMagick is a command line tool you can invoke with the Execute Program operator.

    Hope this helps,

    Rod.


  • rfuentealbarfuentealba Moderator, RapidMiner Certified Analyst, Member, University Professor Posts: 568 Unicorn
    Solution Accepted
    I can.

    Look at this chart please, can RM analyze sequence of images like this? ( each square has an image and they have a sequence like matrix)

    Well... basically all images are glorified sets of numbers. I can make an icon with the German flag with something like this:

    (0,0,0) (0,0,0) (0,0,0)
    (221,0,0) (221,0,0) (221,0,0)
    (255,206,0) (255,206,0) (255,206,0)

    Most analyses as neural networks, however, don't make use of a matrix. Instead, a neural network (I'm not entering into details about convolutional neural networks here) asks for an input like this:

    (0,0,0) (0,0,0) (0,0,0) (221,0,0) (221,0,0) (221,0,0) (255,206,0) (255,206,0) (255,206,0)

    Therefore if by "sequence" you mean "how much are images changed", then I would invent something with time series and the set of numbers or something similar. (This is just an invention).

    This means that RapidMiner can analyse images but it all depends on what do you need to do with these. The thing is that RapidMiner doesn't have a way to import/export images as content, to the best of my knowledge at least, so you need to decode those into numbers, probably using Python or something that can help you do the destructive part.

    All the best,

    Rod.

Answers

  • Telcontar120Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn
    If this is image processing, then probably you would need to do something with the Deep Learning extension and putting together different layers for taking the windowing as shown in the diagram.
    Brian T.
    Lindon Ventures 
    Data Science Consulting from Certified RapidMiner Experts
  • [Deleted User][Deleted User] Posts: 0 Learner III
    @Telcontar120

    Thank you for the answer.
    yes exactly but the problem is sequences!!!! can RM read that like matrix?

    Regards 
    mbs
  • Telcontar120Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn
    I am definitely not an expert in the capabilities of the current DL extension. @pschlunder is probably better positioned to answer your question.
    Brian T.
    Lindon Ventures 
    Data Science Consulting from Certified RapidMiner Experts
  • [Deleted User][Deleted User] Posts: 0 Learner III
    For this project I have a lot of images ( like previous screen shot), what is your suggestion for gathering all of them in one place to make a data base?

    thank you 
  • [Deleted User][Deleted User] Posts: 0 Learner III
    @rfuentealba

    hi

    thank you very much for the answer :)

    can you tell me about the first question too?
  • [Deleted User][Deleted User] Posts: 0 Learner III
    @rfuentealba

    thank you great answer o:)B)
  • pschlunderpschlunder Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, RMResearcher, Member Posts: 96 RM Research
    Hi,
    I would not necessarily store images in a data base but rather on disk. Typically you keep them in folders, where the name represents the label.

    RapidMiner can interpret a flattened matrix as such. You can uncheck the "Infer input shape" option on the Deep Learning operator and set the shape to "Convolutional Flattened".

    If you need to process sequential data like you're asking for, you can create an example set where the batch column identifies the sequence and an id column where each id value identifies the step in the current batch. For batch and id columns use the given special roles with the "Set Role" operator.

    You're data could look like this:
    Where batch = 1 is one sequence with 10 sequence entries/steps/images and att1 to att5 could be your flattened image vector like @rfuentealba nicely explained.

    If your data is in that format convert it to a tensor using the "ExampleSet to Tensor" operator of the Deep Learning extension (available since version 0.9.3) and create your network inside the "Deep Learning on Tensor" operator.

    Hope this helps,
    Philipp
Sign In or Register to comment.