How to make an actual prediction?

wesselwessel Member Posts: 537 Maven
edited November 2018 in Help
Let's say I have a dataset with 10 data points x0 ... x9, and I wish to predict the value for x11.
Using the windowing operator with window_size 3 and horizon 2 I get the training dataset below.
This dataset is fine, I can use a learner to learn a model, that can forecast the value for x11.

Label x-2 x-1 x-0
-----------------------------
x4 x0 x1 x2
x5 x1 x2 x3
x6 x2 x3 x4
x7 x3 x4 x5
x8 x4 x5 x6
x9 x5 x6 x7

But in order to make a prediction for x11 I not only need a model, I also need the appropriate model input.
This input data point would be:

Label x-2 x-1 x-0
-----------------------------
? x7 x8 x9
How do I make Rapid Miner generate this input data point?
Of course it is possible to generate this data point by hand, but in a more realistic scenario my dataset will have many more attributes, and my window_size will be much bigger,
so this means I will have to fill my input data point with many more values.

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi,
    well, in an online scenario you will always have a different input than in an offline learning scenario. But applying this online in an productive enviroment needs many other changes, so that forming a correct example set should not be the biggest problem, is it?

    Greetings,
      Sebastian
  • wesselwessel Member Posts: 537 Maven
    Actually my input in the online scenario is exactly the same as in the offline scenario.

    I'm predicting the weather near my house, not sure if you call this "productive" :P

    I manage to create the input I need, but its rather cumbersome.

    Also a prediction can be rather insightful, if its producing all 0's for tomorrows temperature something is wrong! :P
    This find led me to changing some parameters, so it actually makes a more sensible predictions now.
  • Viktor_MeyerViktor_Meyer Member Posts: 6 Contributor II
    Could someone please post a process, that predicts the value for the last row + 1 !
    If got a data set with values for each month, from january 2000 till december 2009 and i want an actual result for january 2010 !
    Thanks
  • jetonjeton Member Posts: 3 Contributor I
    I have a similar problem and I created a new topic about it.

    I have an Excel sheet with positive values from 0 to 9 ordered by date and I want to predict the number for the next day. I used Windowing, Sliding Window Validation, Support Vector Machine and I came closely to the result in every row but in the LAST row (where the prediction label is) it outputs 0 and it doesn't "predict".
    I tried with another sheet with last date + 1 with the number that had to be "predicted" and the model works ok but for last_date + 2, again... 0!!!
    What am I doing wrong?

    Thanks in advance
Sign In or Register to comment.