Clarification Plz

TensorFlow General Discussion
, ,
1

Greetings,

I hope this message finds you well.

I recently reviewed the time series analysis performed at the following link: “https://www.tensorflow.org/tutorials/structured_data/time_series.”

I have a specific question regarding the methodology outlined in the tutorial, particularly concerning a window with an input length of 24, an output length of 24, and a shift of 1. After training, validating, and testing the model, I am curious about the process of forecasting an unknown series using the provided input data.

In the tutorial, it is mentioned that 19 features, including the output label, were utilized. My query pertains to the feasibility of forecasting when only 18 features are available, excluding the output label data. This is particularly relevant because the input incorporates the temperature (output label) series.

I appreciate your assistance in clarifying this matter.

Thanks & Regards

3

Hi @subramaniyam_cabila,

Sorry for the delayed response. The time series forecasting tutorial given in the link as you mentioned, generates a model with a window size of 24, means it takes 24 past data points as input and predicts the next 24 values. The 19 features used include the actual temperature (output label).

Related to your query: feasibility of forecasting when only 18 features are available, excluding the output label data: It is feasible by using AutoRegressive Forecasting particularly when past input values of the predicted data is not available prior. Please refer to the link1, link2 for more information.

Thank You