MAE for every time-step using multistep modules for timeseries forecasting

TensorFlow General Discussion
1

Hello, I have been following the next example : Time series forecasting  |  TensorFlow Core
I am now implementing the multistep LSTM model and trying to compare its performance to a singlestep model. Therefore i was wondering if it is possible to obtain the MAE for every timestep prediction. Indeed, multi_lstm_model.evaluate averages the MAE over all time steps and i cannot find an argument for model.evaluate that does this.
Thank you very much

3

Hi @Bruno_Hernanz ,

To obtain the MAE for every timestep prediction in a multistep LSTM model:

  • Use model.predict to get the predicted values.
  • Iterate over timesteps and calculate MAE for each timestep using tf.keras.metrics.mean_absolute_error .
  • Analyze the results.

Additionally You can refer this documentations for better understanding .
Thank You