tf.contrib.timeseries.WholeDatasetInputFn

Class WholeDatasetInputFn

Defined in tensorflow/contrib/timeseries/python/timeseries/input_pipeline.py.

Supports passing a full time series to a model for evaluation/inference.

Note that this TimeSeriesInputFn is not designed for high throughput, and should not be used for training. It allows for sequential evaluation on a full dataset (with sequential in-sample predictions), which then feeds naturally into predict_continuation_input_fn for making out-of-sample predictions. While this is useful for plotting and interactive use, RandomWindowInputFn is better suited to training and quantitative evaluation.

Methods

__init__

__init__(time_series_reader)

Initialize the TimeSeriesInputFn.

Args:

• time_series_reader: A TimeSeriesReader object.

__call__

__call__()

Call self as a function.

create_batch

create_batch()

A suitable input_fn for an Estimator's evaluate().

Returns:

A dictionary mapping feature names to Tensors, each shape prefixed by [1, data set size] (i.e. a batch size of 1).