Creates a dataset of sliding windows over a timeseries provided as array.
View aliases
Main aliases
tf.keras.preprocessing.timeseries_dataset_from_array(
data,
targets,
sequence_length,
sequence_stride=1,
sampling_rate=1,
batch_size=128,
shuffle=False,
seed=None,
start_index=None,
end_index=None
)
Used in the notebooks
Used in the tutorials |
---|
This function takes in a sequence of data-points gathered at equal intervals, along with time series parameters such as length of the sequences/windows, spacing between two sequence/windows, etc., to produce batches of timeseries inputs and targets.
Returns |
---|
A tf.data.Dataset
instance. If targets
was passed, the dataset yields
tuple (batch_of_sequences, batch_of_targets)
. If not, the dataset yields
only batch_of_sequences
.
Example 1:
Consider indices [0, 1, ... 98]
.
With sequence_length=10, sampling_rate=2, sequence_stride=3
,
shuffle=False
, the dataset will yield batches of sequences
composed of the following indices:
First sequence: [0 2 4 6 8 10 12 14 16 18]
Second sequence: [3 5 7 9 11 13 15 17 19 21]
Third sequence: [6 8 10 12 14 16 18 20 22 24]
...
Last sequence: [78 80 82 84 86 88 90 92 94 96]
In this case the last 2 data points are discarded since no full sequence can be generated to include them (the next sequence would have started at index 81, and thus its last step would have gone over 98).
Example 2: Temporal regression.
Consider an array data
of scalar values, of shape (steps,)
.
To generate a dataset that uses the past 10
timesteps to predict the next timestep, you would use:
input_data = data[:-10]
targets = data[10:]
dataset = timeseries_dataset_from_array(
input_data, targets, sequence_length=10)
for batch in dataset:
inputs, targets = batch
assert np.array_equal(inputs[0], data[:10]) # First sequence: steps [0-9]
# Corresponding target: step 10
assert np.array_equal(targets[0], data[10])
break
Example 3: Temporal regression for many-to-many architectures.
Consider two arrays of scalar values X
and Y
,
both of shape (100,)
. The resulting dataset should consist samples with
20 timestamps each. The samples should not overlap.
To generate a dataset that uses the current timestamp
to predict the corresponding target timestep, you would use:
X = np.arange(100)
Y = X*2
sample_length = 20
input_dataset = timeseries_dataset_from_array(
X, None, sequence_length=sample_length, sequence_stride=sample_length)
target_dataset = timeseries_dataset_from_array(
Y, None, sequence_length=sample_length, sequence_stride=sample_length)
for batch in zip(input_dataset, target_dataset):
inputs, targets = batch
assert np.array_equal(inputs[0], X[:sample_length])
# second sample equals output timestamps 20-40
assert np.array_equal(targets[1], Y[sample_length:2*sample_length])
break