JoaquinAmatRodrigo / skforecast

Time series forecasting with machine learning models
https://skforecast.org
BSD 3-Clause "New" or "Revised" License
992 stars 113 forks source link

RNN - create_and_compile_model #713

Closed AVPokrovsky closed 1 week ago

AVPokrovsky commented 2 weeks ago

Hi, when I set up create_and_compile_model parameters recurrent_units and dense_units, should I consider someway my lags and steps values? Is there any rules or recommendations?

FernandoCarazoMelo commented 1 week ago

Hi @AVPokrovsky ,

The create_and_compile_model function creates all the logic for your RNN forecaster model.

lags and steps are independent of recurrent_units and dense_units. The general rule is that the more steps you want to predict, the more model parameters you need. Both recurrent_units and dense_units increase the number of parameters. You need to iterate by examining the model's learning curve and checking for overfitting, as shown in the documentation.

AVPokrovsky commented 1 week ago

Ок