Hi Tanmoy Chatterjee
Addressing your first query–
predict – It is used to compute the network’s output by processing the entire input sequence at once and returns the corresponding sequence of predictions as output.
- The “LSTM” network state is updated between predictions within the input sequence, but the state is not returned or maintained after the function call.
predictAndUpdateState - This function is used to make a prediction and return the updated network state.
- This allows you to maintain and manage the network state across multiple calls to the function, which is crucial when generating predictions in a loop, where each prediction becomes the input for the next time step.
Addressing your second query –
- Ensure the following if you do not have test set and would like to improve the prediction of the network.
- Adjust the hyperparameters of the “LSTM”, such as the number of layers, the number of units in each layer, the learning rate, and batch size.
- To prevent overfitting, consider adding dropout or L2 regularization to your “LSTM” layers.
- Try to ensemble with other forecasting models like “ARIMA”. Use ARIMA to make initial forecasts and then use the LSTM to model the error of the ARIMA predictions (learning from the past errors).
- If you are having test data, “xTest”, it is appropriate to use it with “predict” function, to evaluate the performance of the model after training the network.
- When using “predict”, the network weights are not updated based on the test set; you are only updating the internal state of the network to make predictions based on the learned parameters.
Refer to the following links for further guidance on the “predict” and “predictAndUpdateState”.
predict –
predictAndUpdateState – https://www.mathworks.com/help/releases/R2020a/deeplearning/ref/predictandupdatestate.html
