Need help for comparing TensorFlow guide with MATLAB tutorial for time series forecasting

General Discussion
, ,
#1

Hello! I have 2 questions regarding TensorFlow guide on time series forecasting

  1. It is stated on the guide that the “window” shape of data should be (batch, time, features). Since fit, evaluate and predict methods of tf.keras.Sequential can handle numpy arrays for x and y inputs, do these inputs’ shape also in (batch, time, features)?

  2. In this MATLAB example, there are 2 types of forecasting discussed: Open-loop Forecasting and Close-loop Forecasting. Is TensorFlow’s guide also cover these topics? If it is, which part is Open-loop Forecasting and which part is Closed-loop Forecasting?

Thank you in advance

#3

Hi @Muhammad_Robith

Welcome to the TensorFlow Forum!

  1. Yes, inputs shapes can be in (batch, time, features) format for fit , evaluate and predict methods of tf.keras.Sequential.
  2. TensorFlow doesn’t explicitly use the terms “open-loop” and “closed-loop” forecasting, but it covers both concepts: Single-step and Multi-step models in the guide approach follows the Open-loop Forecasting and likewise you can implement closed-loop forecasting by using a recurrent architecture like LSTM (Long Short-Term Memory).

Thank you.