Time Series Fundamentals: Exercise 4

[maksytka03]

Hi all. I have a problem with exercise 4, chapter 10. I've sort of figured out how to make a dataset, but don't seem to understand what the sequence_length attribute does in the tf.keras.utils.timeseries_dataset_from_array function.

My way of doing things:
train_dataset = tf.keras.utils.timeseries_dataset_from_array(train_windows, targets=train_labels, sequence_length=WINDOW_SIZE) test_dataset = tf.keras.utils.timeseries_dataset_from_array(test_windows, targets=test_labels, sequence_length=WINDOW_SIZE)

Creates two datasets, which I train model_1 on.

After that, I try to get the results of the model using the evaluate_preds function we've made in the course.

The code errors out, so I figured I can try to use the following code to make it possible to use the evaluate_preds function:

  dataset_predictions = []
  ground_truth = []
  for batch in test_dataset:
    batch_pred = model_1_dataset.predict(batch)
    dataset_predictions.append(batch_pred)
    ground_truth.append(batch)
  
  dataset_predictions = tf.concat(dataset_predictions, axis=0).numpy()
  ground_truth = tf.concat(ground_truth, axis=0).numpy()

But I get the following error:
ValueError: Layer "model_1_dense_dataset" expects 1 input(s), but it received 2 input tensors. Inputs received: [<tf.Tensor 'IteratorGetNext:0' shape=(32, 7, 7) dtype=float64>, <tf.Tensor 'IteratorGetNext:1' shape=(32, 1) dtype=float64>]

Any help on how to do the exercise properly will be appreciated.