Work on series data

silbgbsilbgb Member Posts: 2 Contributor I
edited November 2018 in Help
Hi there!

I'm trying to use some analysis with series. I have a file of examples, one example per row. Each row has a different number of attributes, like:

att1-0, att2-0, att1-1, att2-1, att1-2, att2-2, att1-3, att2-3
att1-0, att2-0, att1-1, att2-1

i.e., the example's length depends on the time, actually frames.

How can I process each? or better, transform them in order to have the same length or windowing, something like the DTWDistance, that is in RapidMiner, but I don't know how to use it, find an example or documentation.

I really appreciate if you can help me to find where I can learn series.

Thanks a lot,

Silvia

Answers

  • qwertzqwertz Member Posts: 130 Contributor II
    Hi Silvia,

    I really would like to help you but to be honest I don't understand your question.

    -What type of data do you have (stocks, weather, traffic volume)? It seems strange to me to have a dataset where the number of attributes varies. Or would you just like to transpose the data, i.e. switch rows and coloums or to put it in a Rapidminer wording: examples and attributes?
    -What is the desired format you would like to have?

    PS: I would recommend to open a new thread for this topic. Though it might have similarities it appears to be another question to me.


    Kind regards
    Sachs
  • silbgbsilbgb Member Posts: 2 Contributor I
    Hi Sachs!

    Thank you for  your reply ;)

    I have data that represent gestures. Each gesture has several poses. Each pose is represented by 15 points. But the gestures take differents times each one. Something similar to trying to recognize voice, the same word can take different time for different people.
    The DTW (Dynamic Time Warping) is a tool for time series analysis, for measuring similarity between two sequences wihich may vary in time or speed.

    I'm know in rapidminer, that's great!! for introductory the videos in the web have been very useful. But now for series, I can't find documentation to help me. Do you know something?

    Best,

    Silvia
  • wesselwessel Member Posts: 537 Maven
    Right, a hand gesture data set.
    Can you paste a few rows of data?

    I imagine your data looks the following:
    You measure 3 (or 15) points on the hand: a, b, and c.

    Now at every millisecond you measure the position of these points in 1D or in 3D?
    So a single hand gesture is represented by a large vector containing 300 data points.
    a-1, a-2, a-3, ... a-100
    b-1, b-2, b-3, ... b-100
    c-1, c-2, c-3, ... c-100

    This is not really time series data to be honest.
    Since stuff like trends or moving averages are not really important.
    But you can use time series techniques and all the standard classification techniques.


  • qwertzqwertz Member Posts: 130 Contributor II

    Hi Silvia,

    I think wessel's idea to have a small sample of your data set would indeed be helpful. Another questions that has remained is what you want to get out of the data...

    Cheers
    Sachs
Sign In or Register to comment.