1. Learn
    2. /
  2. Courses
    3. /
  3. Intermediate Deep Learning with PyTorch

Exercise

Sequential Dataset

Good job building the create_sequences() function! It's time to use it to create a training dataset for your model.

Just like tabular and image data, sequential data is easiest passed to a model through a torch Dataset and DataLoader. To build a sequential Dataset, you will call create_sequences() to get the NumPy arrays with inputs and targets, and inspect their shape. Next, you will pass them to a TensorDataset to create a proper torch Dataset, and inspect its length.

Your implementation of create_sequences() and a DataFrame with the training data called train_data are available.

Instructions

100 XP
  • Call create_sequences(), passing it the training DataFrame and a sequence length of 24*4, assigning the result to X_train, y_train.
  • Define dataset_train by calling TensorDataset and passing it two arguments, the inputs and the targets created by create_sequences(), both converted from NumPy arrays to tensors of floats.