Lines Matching refs:features
27 or more floating point values for each time) along with any exogenous features,
39 features may have any shape, but are likewise associated with a timestep. Times
42 features (i.e. times may be missing, but given that a time is specified, every
119 features for the model to use while making the predictions. Values must
134 features = {
140 features.update(exogenous_features)
146 for key, values in features.items():
234 has shape [series length x number of features].
235 In any case, `VALUES` and any exogenous features must have their shapes
256 # Remove the batch dimension from all features
257 features = {key: numpy.squeeze(value, axis=0)
260 x=features,
261 # The first dimensions of features are the series length, since we have
271 features = estimator_lib.inputs.numpy_input_fn(
279 for feature_name, feature_value in features.items()}
493 features = {}
497 features[column_name] = values[0]
499 features[column_name] = array_ops.stack(values, axis=1)
500 return features
508 features):
514 features: A dictionary mapping from feature keys to `tf.FixedLenFeature`
516 `TrainEvalFeatures.VALUES` (floating point vector) features.
518 ValueError: If required times/values features are not present.
520 if feature_keys.TrainEvalFeatures.TIMES not in features:
523 if feature_keys.TrainEvalFeatures.VALUES not in features:
526 self._features = features
535 serialized=examples, features=self._features)
543 """Creates chunked Tensors from times, values, and other features.
549 A tuple of (features, targets):
550 features: A dictionary with `TrainEvalFeatures.TIMES` and
553 size x window length x number of features]. Any other features will
600 features = self._reader.read_full()
603 for feature_name, feature_value in features.items()},
649 False, the window and associated features are instead sorted so that
692 features = self._reader.read()
704 features,
791 vector of length [number of features].
794 features] (multivariate).
797 size x series length x number of features].
798 In any case, `VALUES` and any exogenous features must have their shapes
803 A dictionary with features normalized to have shapes prefixed with [batch
808 features are missing.
810 features = {key: numpy.array(value) for key, value in data.items()}
811 if (feature_keys.TrainEvalFeatures.TIMES not in features or
812 feature_keys.TrainEvalFeatures.VALUES not in features):
813 raise ValueError("{} and {} are required features.".format(
816 times = features[feature_keys.TrainEvalFeatures.TIMES]
817 for key, value in features.items():
820 ("All features must have their shapes prefixed by the shape of the"
825 if not features[feature_keys.TrainEvalFeatures.VALUES].shape: # univariate
827 features[feature_keys.TrainEvalFeatures.VALUES] = features[
829 elif len(features[feature_keys.TrainEvalFeatures.VALUES].shape) > 1:
833 " ([number of features]) since '{}' does not "
837 features[feature_keys.TrainEvalFeatures.VALUES].shape))
839 features = {key: value[None, None, ...] for key, value in features.items()}
841 if len(features[feature_keys.TrainEvalFeatures.VALUES]
844 features[feature_keys.TrainEvalFeatures.VALUES] = features[
846 elif len(features[feature_keys.TrainEvalFeatures.VALUES].shape) > 2:
850 " ([series length, number of features]) since '{}' does not "
854 features[feature_keys.TrainEvalFeatures.VALUES].shape))
856 features = {key: value[None, ...] for key, value in features.items()}
857 elif len(features[feature_keys.TrainEvalFeatures.TIMES]
865 if features[feature_keys.TrainEvalFeatures.TIMES].shape[0] != 1:
867 return features
