Time Series Forecasting using DeepAR and GluonTS

Поделиться
HTML-код
  • Опубликовано: 8 ноя 2024
  • НаукаНаука

Комментарии • 52

  • @mattpaterson5615
    @mattpaterson5615 3 года назад +3

    This is one of the best "How-to-use-this-ML-Algorithm" tutorials on the web re:DeepAR. Thank you for helping me make my 'hello_deepar_world' model!

  • @kamaldhungana6677
    @kamaldhungana6677 9 месяцев назад +1

    Where is the actual forecast? You explained the training and testing parts very nicely. For future prediction, how do we set up the input meta data so that we make prediction for another 10 days?

  • @jordon603
    @jordon603 4 года назад +3

    Hi, thanks for the video. Looking forward to multivariate forecasting

  • @ashwinipadhy6397
    @ashwinipadhy6397 3 года назад +1

    Thanks for sharing this and great explanation.I have few questions:
    1. GluonTS is open source and I can use it in my ec2 or like deep AR I have to use in sagemaker only.
    2. Like DeepAR do we have option to consider related time series and item metadata(city_id,product_name etc), holiday etc.
    3. Do we have to do missing value treatment before providing dataset for prediction or gluonts handles that internally.

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      You can use both DeepAR and other GluonTS models in ec2. Sagemaker just provides convinient way of accessing DeepAR but it can be used outside as well
      Yes you can provide multivariate time features as well as multiple time series as well. I have covered it in detailed example in my multiple time series video
      Yes better null values are handled in program. I head some latest updates to DeepAR does not need it but have not tried it

  • @jayapalsadasivan2003
    @jayapalsadasivan2003 3 года назад

    Hi Sir! You have made the best tutorial video ever in this topic. Please make more videos on Machine Learning and similar deep learning concepts. Thank you so much. Regards & best wishes.

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад +1

      Jayapal.. I have few on deep learning as well. You can check my NLP playlist if you are interested

    • @jayapalsadasivan2003
      @jayapalsadasivan2003 3 года назад

      @@AIEngineeringLife definitely sir!

  • @hemantnyadav
    @hemantnyadav Год назад

    Hi, this was a very good video but is it multivariate time series? As there are many variables you selected for estimating Appliance usage... Also in this, you are not using any lag features? Can we accommodate lag feature for better predictions?

  • @karthidhoni
    @karthidhoni 4 года назад +1

    Nice explanation. Looking forward for the video on multivariate analysis.

    • @AIEngineeringLife
      @AIEngineeringLife  4 года назад +1

      In few hours, multiple time series must be live. Multi variate will be over the weekend ideally with VAR

    • @karthidhoni
      @karthidhoni 4 года назад

      @@AIEngineeringLife That would be really great and if possible please do share the colab notebook for DeepAR

    • @AIEngineeringLife
      @AIEngineeringLife  4 года назад +1

      @@karthidhoni yes will upload in my git repo soon

    • @karthidhoni
      @karthidhoni 4 года назад

      @@AIEngineeringLife Thanks that would be helpful

    • @malleswararaomaguluri6344
      @malleswararaomaguluri6344 3 года назад

      Deepar and gluonts will work on CPU. How much memory required to run for 1 day ahead forecast using this. Freq is 15 min data I m using.

  • @goelnikhils
    @goelnikhils 2 года назад

    Amazing Video on Demand Forecasting. Thanks a lot .

  • @venkateswarluvobbani8658
    @venkateswarluvobbani8658 3 года назад +4

    Hi Sir, Can you please share the link for code? @AIEngineering
    Thanks in advance!!

  • @harivilasreddy
    @harivilasreddy 3 года назад +3

    can you make a video on multivariate timeseries forecasting using DeepAR and GluonTS?

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      Hari.. I will try with DeepAR but I do have videos on multivariate with other time series models like VAR, LSTM and others

    • @harivilasreddy
      @harivilasreddy 3 года назад

      @@AIEngineeringLife I used use_feat_dynamic_real=True and added extra feature to train dataset. Successfully built the predictor and same used for prediction also. But once prediction generator makes my colab crashing. This one is not happening with use_feat_dynamic_real=False. Do you have any idea?

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      @@harivilasreddy That's surprising. Are you using gpu or cpu?

    • @harivilasreddy
      @harivilasreddy 3 года назад

      @@AIEngineeringLife GPU

    • @harivilasreddy
      @harivilasreddy 3 года назад

      @@AIEngineeringLife can we have a discussion it if you feel free?

  • @alteshaus3149
    @alteshaus3149 3 года назад +1

    Thank you for this great video! I would love to see more from GluonTS, could you show some more usage of TimeSeries Modelling with it?
    regards from Austria

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      I have 2 videos on GluonTs. You can check my time series course playlist here - github.com/srivatsan88

    • @yaswanthbangaru9116
      @yaswanthbangaru9116 3 года назад

      @@AIEngineeringLife any plans on doing a session related to transformers use for time series data?

  • @antivistrock702
    @antivistrock702 4 года назад +1

    Thank you for your video! Have you posted a link of the colab notebook from this video?

    • @AIEngineeringLife
      @AIEngineeringLife  4 года назад +1

      Not yet.. Will do it in a day or 2

    • @КириллГорбань-к2г
      @КириллГорбань-к2г 4 года назад +1

      @@AIEngineeringLife Waiting for it. Please tag me to check it out. I'd also ask to comment collab code to fully understand the concept. Thanks a lot for this video.

    • @AIEngineeringLife
      @AIEngineeringLife  4 года назад

      @@КириллГорбань-к2г My code is already in my git repo - github.com/srivatsan88/End-to-End-Time-Series
      I would try to comment going forward but typically I get only weekend to do video and if I try to cover all aspects it would really take lot of time. Hope you understand

  • @SanataniAryavrat
    @SanataniAryavrat 3 года назад

    Good! well explained tutorial. Thanks.

  • @DmitryPesegov
    @DmitryPesegov 23 дня назад

    Wouldn't optimisation just learn "variance"-variable in the denominator of Gaussian so huge so that almost all possible targets values from that learned Gaussians will give probabilities eq to 1 ?

  • @shaythuramelangkovan5800
    @shaythuramelangkovan5800 3 года назад

    How do we configure it so that it only looks at business days ?? My data set has gaps where Saturday and Sunday isn’t counted. Can I get help w that ?

  • @falaknaaz7139
    @falaknaaz7139 3 года назад

    Hi, thanks for the great video. Can gluonts be used for data with timestamps instead of date? if yes how can we modify the code for the same?

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      When you say timestamp it is data + time right. In that case no changes are required you can feed it directly instead of date

    • @falaknaaz7139
      @falaknaaz7139 3 года назад

      @@AIEngineeringLife the thing is I have battery data with timestamps in seconds starting from 00:00:00 when I feed this to the network it shows an error..

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      @@falaknaaz7139 If you have only time then you can add some dummy date like todays date before it and make it date time. Only time these models will not take

    • @falaknaaz7139
      @falaknaaz7139 3 года назад

      ​@@AIEngineeringLife I have done the same but wanted to know if there is an alternative. thanks!

    • @falaknaaz7139
      @falaknaaz7139 3 года назад

      @@AIEngineeringLife Sir is Gluonts good for forecasting voltage values? I have data with voltage, current, and temperature values and I fed only voltage and time values while training, I wanted your suggestion in this and should I feed the other parameters as well to improve the performance?

  • @AmitBarik-e3w
    @AmitBarik-e3w Год назад

    Cud u pls add the inferencing part to it. I meant how to forecast for future dates??

  • @Titisaintmand
    @Titisaintmand 2 года назад

    Hello, I've written your code, but there is many errors when runs. Could you let your code ? Or can I send my code to see where there is problem ? Thanks very much

  • @alishametkari5137
    @alishametkari5137 2 года назад

    How to use time_feature that it has?

  • @roycesamuel6947
    @roycesamuel6947 3 года назад

    Hey! Could you share this source.

  • @alishametkari5137
    @alishametkari5137 2 года назад

    How to use metadata and related time series with gluonts models, specially deepar?? Can anyone help?

  • @malleswararaomaguluri6344
    @malleswararaomaguluri6344 3 года назад

    How much memory required to run this. How much run time it's required in CPU machine.

    • @AIEngineeringLife
      @AIEngineeringLife  3 года назад

      It depends on data size and columns. If on cpu you can run with smaller batches and training can take lot of time. I have 64gb memory but memory is not used more than 8gb. Each epoch takes lot of time though. Have better cores so matrix operations can be run in parallel

  • @vibesnovibes6320
    @vibesnovibes6320 3 года назад

    Anyone else facing issues installing mxnet? I am facing a build numpy error :(

  • @simple_sci
    @simple_sci 9 месяцев назад

    ruclips.net/video/4Mkne5AmlgY/видео.html