Time series data prediction problems

Hello guys, I am doing a project which uses the input signal of a power amplifier to predict its output signal. Both signals are sinusoidal signals. My plan is to build a seq2seq LSTM model to complete the task. But I have some questions which I need some explainations.

First, I want to use time_series_split method to train and validate the model, however, this is my first time doing it, I am wondering if it’s possible to implement this method with a customize dataset and dataloader, and how to do that as well?

Second, I will have hundreds of sinusoidal signals with different amplitude and frequency as input dataset, does it impove the model performance if I use batches?

Question One : Yes can you implement it with a customized dataset. You can check out Writing Custom Datasets, Dataloaders and Transform on how to create custom datasets.

Question Two: Yes, training using batches can have a direct impact on accuracy.

1 Like