Topic Thread

Expand all | Collapse all

Time Series Pattern Identification

  • 1.  Time Series Pattern Identification

    Posted Tue February 09, 2021 11:08 AM
    I hope this is the right platform to ask this, but I have an idea of training a model with time series data, and giving it pattern examples so that it can use that for prediction. I have tried training it with a sample of data and realized that it will take an awful lot of time to train it. Any ideas of how to get this done using Tensorflow and Pandas?

    ------------------------------
    Joseph Lubinda
    ------------------------------


  • 2.  RE: Time Series Pattern Identification

    Posted Wed February 10, 2021 11:16 AM
    Hello Joseph,

    Can you be more clear please? ( which langage python or R?)
    Do you know facebook powerful library called fbprophet?
    It is a good library for time serie forecast ( very speed and very performant).
    With pandas this is not possible, but tensorflow takes sometime a lot of time depending on the type of model :  classic model like arima , sarima are speed  and accurate but modern model like CNN , Deep AR, will take a lot of time and are not sometime more accurate than classic.

    See  ressoures(  the first 3 ressourses are for facebook fb prophet) :
          https://facebook.github.io/prophet/docs/quick_start.html#python-api
          https://towardsdatascience.com/implementing-facebook-prophet-efficiently-c241305405a3
          https://towardsdatascience.com/time-series-analysis-in-python-an-introduction-70d5a5b1d52a
          https://machinelearningmastery.com/time-series-forecasting-methods-in-python-cheat-sheet/
          https://machinelearningmastery.com/xgboost-for-time-series-forecasting/


    ------------------------------
    ismael amadou hassane
    ------------------------------