View source on GitHub
|
A time series parser for feeding Numpy arrays to a TimeSeriesInputFn.
tf.contrib.timeseries.NumpyReader(
data, read_num_records_hint=4096
)
Avoids embedding data in the graph as constants.
Args | |
|---|---|
data
|
A dictionary mapping feature names to Numpy arrays, with two
possible shapes (requires keys TrainEvalFeatures.TIMES and
TrainEvalFeatures.VALUES): Univariate; TIMES and VALUES are both
vectors of shape [series length] Multivariate; TIMES is a vector of
shape [series length], VALUES has shape [series length x number of
features]. In any case, VALUES and any exogenous features must have
their shapes prefixed by the shape of the value corresponding to the
TIMES key.
|
read_num_records_hint
|
The maximum number of samples to read at one time, for efficiency. |
Methods
check_dataset_size
check_dataset_size(
minimum_dataset_size
)
Raise an error if the dataset is too small.
read
read()
Returns a large chunk of the Numpy arrays for later re-chunking.
read_full
read_full()
Returns Tensor versions of the full Numpy arrays.