Hitachi Vantara Pentaho Community Forums
Results 1 to 2 of 2

Thread: Can anyone help me with the prime process and incremental learning?

  1. #1
    Join Date
    Mar 2016
    Posts
    1

    Post Can anyone help me with the prime process and incremental learning?

    Hi! I am new user of WEKA, and I love it because of its good performance and usability.

    However, I recently have a problem with time series forecasting.

    The thing is, I want to use WEKA to construct a incremental learning platform. At first I was thinking of using sliding windows to realize it, because what we want is to update the model for every several inputs (like 10) and we don't need to update the model immediately when we have a new input.

    And then, I notice the function primeForecasterIncrementally(), and I also notice that if I call this function for every new instance, the outcome would be different. However, I looked up your forum, Mark said WEKA does not support incremental learning at this moment, all training is batch-based even if the underlying learner can be trained incrementally.

    So my question is,
    1) Even if the training is batch-based, can I apply the primeForecasterIncrementally() to satisfy my needs.
    2) What is priming process doing? Why do we need to call primeForecaster() after we have called buildForecaster()?

    I would appreciate if anyone can help me.

  2. #2
    Join Date
    Aug 2006
    Posts
    1,741

    Default

    The forecasting environment is indeed a batch learning processes. Priming is the process of flattening a number of time series data points into a single instance (essentially windowed - that is what the "lagged" variables are) that can then be input into the underlying propositional regression algorithm to produce a prediction. So this process is used at prediction time, and is the reason why there needs to be x (where x is as long as the longest lagged variable) historical data points immediately prior (in time) to the first forecasted time unit in order to kick off the closed loop forecasting process.

    I'm afraid that, at present, the forecasting environment is not set up to handle incremental training. There are underlying incremental regression algorithms that could be used for this - e.g. SGD - but the forecasting code would need considerable changes to work incrementally.

    Cheers,
    Mark.

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
Privacy Policy | Legal Notices | Safe Harbor Privacy Policy

Copyright © 2005 - 2019 Hitachi Vantara Corporation. All Rights Reserved.