TensorFlow Tutorial #23 Time-Series Prediction




[ad_1]

How to predict time-series data using a Recurrent Neural Network (GRU / LSTM) in TensorFlow and Keras. Demonstrated on weather-data.

https://github.com/Hvass-Labs/TensorFlow-Tutorials

This tutorial has been updated to work with TensorFlow 2.1 and possibly later versions.

Source


[ad_2]

Comment List

  • Hvass Laboratories
    December 4, 2020

    The only understandable tutorial on this topic I've found. No one else knows to explain the reasons why they do what they do I guess, so that I can actually fucking learn it, thank you!!!

  • Hvass Laboratories
    December 4, 2020

    people from Denmark and Norway are very intelligent. They have monster brains

  • Hvass Laboratories
    December 4, 2020

    I want to classify anomaly detection using RNN keras.tf but I have a problem where the accuracy value increases but the val_accuracy value does not change and just remains constant at 50%. this is my complete code available on google colab https://colab.research.google.com/drive/1saoNuCxj08JCxZ_7taIjhp8sJEIV-T5U?usp=sharing //

  • Hvass Laboratories
    December 4, 2020

    With this implementation the predictor variables are available at all time steps in the forecast, but the forecasted variables are not, right?

  • Hvass Laboratories
    December 4, 2020

    Can I use this for a dataset with less than 20 records ?

  • Hvass Laboratories
    December 4, 2020

    I have paid 200 usd to study these lessons at school. Now, I can learn tl on this channel and they are free.
    God bless you.

  • Hvass Laboratories
    December 4, 2020

    Still very good material, and going much more in depth than other sources. Kudos!

  • Hvass Laboratories
    December 4, 2020

    Hi Magnus,

    Excellent tutorials on TF TS. I am not able to find the weather data for Denmark in the link NCDC as mentioned. Also the import inclusion – import weather – is not having any effect. No download / import is happening. Please help me on how to get the weather data. Regards, KM

  • Hvass Laboratories
    December 4, 2020

    thanks for sharing. Please upload more content!

  • Hvass Laboratories
    December 4, 2020

    I’m having error after running In83 as profiler not running error
    Can you help me?

  • Hvass Laboratories
    December 4, 2020

    So glad I found this video! A question if I may, does anyone know how to only use the last 60 days to predict the next day, rather than the entire prior data?

  • Hvass Laboratories
    December 4, 2020

    Good tutorial. Aweful results.

  • Hvass Laboratories
    December 4, 2020
  • Hvass Laboratories
    December 4, 2020

    Thanks for the tutorial. I may not hear so many "here(s)" anywhere else except here.

  • Hvass Laboratories
    December 4, 2020
  • Hvass Laboratories
    December 4, 2020

    Many thanks for this video Magnus, learned a lot – callbacks were amazing! One observation re the custom loss function. You've built that to allow a warm up period, statistically makes a lot sense. I tried implementing a warm up period in a different way in my own model, using sample weights by adding this code to model fit "sample_weight=np.concatenate((np.zeros(20), np.ones(len(x_train_scaled)-20)), axis=0)". This could be a more robust way of doing it, as it doesn't seem to cause NaNs when input dimension or batch size changes.

  • Hvass Laboratories
    December 4, 2020

    if you're using the past data to predict future weather, why is the y_data also shifted as the x_data did

  • Hvass Laboratories
    December 4, 2020

    good tutorial, but have you excluded your y variable from your x_data?

  • Hvass Laboratories
    December 4, 2020

    Great tutorial my friend

  • Hvass Laboratories
    December 4, 2020

    Hello Erik, Can we fill out the missing data by using LSTM ???? and How ???

Write a comment