Source code for gluonts.model.deepstate._estimator

# Copyright 2018 Amazon.com, Inc. or its affiliates. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License").
# You may not use this file except in compliance with the License.
# A copy of the License is located at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# or in the "license" file accompanying this file. This file is distributed
# on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
# express or implied. See the License for the specific language governing
# permissions and limitations under the License.

from functools import partial
from typing import List, Optional

import numpy as np
from mxnet.gluon import HybridBlock
from pandas.tseries.frequencies import to_offset

from gluonts.core.component import validated
from gluonts.dataset.common import Dataset
from gluonts.dataset.field_names import FieldName
from gluonts.dataset.loader import (
    DataLoader,
    TrainDataLoader,
    ValidationDataLoader,
)
from gluonts.model.deepstate.issm import ISSM, CompositeISSM
from gluonts.model.predictor import Predictor
from gluonts.mx.batchify import batchify
from gluonts.mx.distribution.lds import ParameterBounds
from gluonts.mx.model.estimator import GluonEstimator
from gluonts.mx.model.predictor import RepresentableBlockPredictor
from gluonts.mx.trainer import Trainer
from gluonts.mx.util import copy_parameters, get_hybrid_forward_input_names
from gluonts.time_feature import (
    TimeFeature,
    norm_freq_str,
    time_features_from_frequency_str,
)
from gluonts.transform import (
    AddAgeFeature,
    AddObservedValuesIndicator,
    AddTimeFeatures,
    AsNumpyArray,
    CanonicalInstanceSplitter,
    Chain,
    ExpandDimArray,
    RemoveFields,
    SelectFields,
    SetField,
    TestSplitSampler,
    Transformation,
    VstackFeatures,
)

from ._network import DeepStatePredictionNetwork, DeepStateTrainingNetwork

SEASON_INDICATORS_FIELD = "seasonal_indicators"


# A dictionary mapping granularity to the period length of the longest season
# one can expect given the granularity of the time series. This is similar to
# the frequency value in the R forecast package:
# https://stats.stackexchange.com/questions/120806/frequency-value-for-seconds-minutes-intervals-data-in-r
# This is useful for setting default values for past/context length for models
# that do not do data augmentation and uses a single training example per time
# series in the dataset.
FREQ_LONGEST_PERIOD_DICT = {
    "M": 12,  # yearly seasonality
    "W": 52,  # yearly seasonality
    "D": 31,  # monthly seasonality
    "B": 22,  # monthly seasonality
    "H": 168,  # weekly seasonality
    "T": 1440,  # daily seasonality
}


def longest_period_from_frequency_str(freq_str: str) -> int:
    offset = to_offset(freq_str)
    return FREQ_LONGEST_PERIOD_DICT[norm_freq_str(offset.name)] // offset.n


[docs]class DeepStateEstimator(GluonEstimator): """ Construct a DeepState estimator. This implements the deep state space model described in [RSG+18]_. Parameters ---------- freq Frequency of the data to train on and predict prediction_length Length of the prediction horizon cardinality Number of values of each categorical feature. This must be set by default unless ``use_feat_static_cat`` is set to `False` explicitly (which is NOT recommended). add_trend Flag to indicate whether to include trend component in the state space model past_length This is the length of the training time series; i.e., number of steps to unroll the RNN for before computing predictions. Set this to (at most) the length of the shortest time series in the dataset. (default: None, in which case the training length is set such that at least `num_seasons_to_train` seasons are included in the training. See `num_seasons_to_train`) num_periods_to_train (Used only when `past_length` is not set) Number of periods to include in the training time series. (default: 4) Here period corresponds to the longest cycle one can expect given the granularity of the time series. See: https://stats.stackexchange.com/questions/120806/frequency -value-for-seconds-minutes-intervals-data-in-r trainer Trainer object to be used (default: Trainer()) num_layers Number of RNN layers (default: 2) num_cells Number of RNN cells for each layer (default: 40) cell_type Type of recurrent cells to use (available: 'lstm' or 'gru'; default: 'lstm') num_parallel_samples Number of evaluation samples per time series to increase parallelism during inference. This is a model optimization that does not affect the accuracy ( default: 100). dropout_rate Dropout regularization parameter (default: 0.1) use_feat_dynamic_real Whether to use the ``feat_dynamic_real`` field from the data (default: False) use_feat_static_cat Whether to use the ``feat_static_cat`` field from the data (default: True) embedding_dimension Dimension of the embeddings for categorical features (default: [min(50, (cat+1)//2) for cat in cardinality]) scaling Whether to automatically scale the target values (default: true) time_features Time features to use as inputs of the RNN (default: None, in which case these are automatically determined based on freq) noise_std_bounds Lower and upper bounds for the standard deviation of the observation noise prior_cov_bounds Lower and upper bounds for the diagonal of the prior covariance matrix innovation_bounds Lower and upper bounds for the standard deviation of the observation noise batch_size The size of the batches to be used training and prediction. """ @validated() def __init__( self, freq: str, prediction_length: int, cardinality: List[int], add_trend: bool = False, past_length: Optional[int] = None, num_periods_to_train: int = 4, trainer: Trainer = Trainer( epochs=100, num_batches_per_epoch=50, hybridize=False ), num_layers: int = 2, num_cells: int = 40, cell_type: str = "lstm", num_parallel_samples: int = 100, dropout_rate: float = 0.1, use_feat_dynamic_real: bool = False, use_feat_static_cat: bool = True, embedding_dimension: Optional[List[int]] = None, issm: Optional[ISSM] = None, scaling: bool = True, time_features: Optional[List[TimeFeature]] = None, noise_std_bounds: ParameterBounds = ParameterBounds(1e-6, 1.0), prior_cov_bounds: ParameterBounds = ParameterBounds(1e-6, 1.0), innovation_bounds: ParameterBounds = ParameterBounds(1e-6, 0.01), batch_size: int = 32, ) -> None: super().__init__(trainer=trainer, batch_size=batch_size) assert ( prediction_length > 0 ), "The value of `prediction_length` should be > 0" assert ( past_length is None or past_length > 0 ), "The value of `past_length` should be > 0" assert num_layers > 0, "The value of `num_layers` should be > 0" assert num_cells > 0, "The value of `num_cells` should be > 0" assert ( num_parallel_samples > 0 ), "The value of `num_parallel_samples` should be > 0" assert dropout_rate >= 0, "The value of `dropout_rate` should be >= 0" assert not use_feat_static_cat or any(c > 1 for c in cardinality), ( "Cardinality of at least one static categorical feature must be" " larger than 1 if `use_feat_static_cat=True`. But cardinality" f" provided is: {cardinality}" ) assert embedding_dimension is None or all( e > 0 for e in embedding_dimension ), "Elements of `embedding_dimension` should be > 0" assert all( np.isfinite(p.lower) and np.isfinite(p.upper) and p.lower > 0 for p in [noise_std_bounds, prior_cov_bounds, innovation_bounds] ), ( "All parameter bounds should be finite, and lower bounds should be" " positive" ) self.past_length = ( past_length if past_length is not None else num_periods_to_train * longest_period_from_frequency_str(freq) ) self.prediction_length = prediction_length self.add_trend = add_trend self.num_layers = num_layers self.num_cells = num_cells self.cell_type = cell_type self.num_parallel_samples = num_parallel_samples self.scaling = scaling self.dropout_rate = dropout_rate self.use_feat_dynamic_real = use_feat_dynamic_real self.use_feat_static_cat = use_feat_static_cat self.cardinality = ( cardinality if cardinality and use_feat_static_cat else [1] ) self.embedding_dimension = ( embedding_dimension if embedding_dimension is not None else [min(50, (cat + 1) // 2) for cat in self.cardinality] ) self.issm = ( issm if issm is not None else CompositeISSM.get_from_freq(freq, add_trend) ) self.time_features = ( time_features if time_features is not None else time_features_from_frequency_str(freq) ) self.noise_std_bounds = noise_std_bounds self.prior_cov_bounds = prior_cov_bounds self.innovation_bounds = innovation_bounds
[docs] def create_transformation(self) -> Transformation: remove_field_names = [ FieldName.FEAT_DYNAMIC_CAT, FieldName.FEAT_STATIC_REAL, ] if not self.use_feat_dynamic_real: remove_field_names.append(FieldName.FEAT_DYNAMIC_REAL) return Chain( [RemoveFields(field_names=remove_field_names)] + ( [SetField(output_field=FieldName.FEAT_STATIC_CAT, value=[0.0])] if not self.use_feat_static_cat else [] ) + [ AsNumpyArray(field=FieldName.FEAT_STATIC_CAT, expected_ndim=1), AsNumpyArray(field=FieldName.TARGET, expected_ndim=1), # gives target the (1, T) layout ExpandDimArray(field=FieldName.TARGET, axis=0), AddObservedValuesIndicator( target_field=FieldName.TARGET, output_field=FieldName.OBSERVED_VALUES, ), # Unnormalized seasonal features AddTimeFeatures( time_features=self.issm.time_features(), pred_length=self.prediction_length, start_field=FieldName.START, target_field=FieldName.TARGET, output_field=SEASON_INDICATORS_FIELD, ), AddTimeFeatures( start_field=FieldName.START, target_field=FieldName.TARGET, output_field=FieldName.FEAT_TIME, time_features=self.time_features, pred_length=self.prediction_length, ), AddAgeFeature( target_field=FieldName.TARGET, output_field=FieldName.FEAT_AGE, pred_length=self.prediction_length, log_scale=True, ), VstackFeatures( output_field=FieldName.FEAT_TIME, input_fields=[FieldName.FEAT_TIME, FieldName.FEAT_AGE] + ( [FieldName.FEAT_DYNAMIC_REAL] if self.use_feat_dynamic_real else [] ), ), ] )
def _create_instance_splitter(self, mode: str): assert mode in ["training", "validation", "test"] return CanonicalInstanceSplitter( target_field=FieldName.TARGET, is_pad_field=FieldName.IS_PAD, start_field=FieldName.START, forecast_start_field=FieldName.FORECAST_START, instance_sampler=TestSplitSampler(), time_series_fields=[ FieldName.FEAT_TIME, SEASON_INDICATORS_FIELD, FieldName.OBSERVED_VALUES, ], allow_target_padding=True, instance_length=self.past_length, use_prediction_features=(mode != "training"), prediction_length=self.prediction_length, )
[docs] def create_training_data_loader( self, data: Dataset, **kwargs, ) -> DataLoader: input_names = get_hybrid_forward_input_names(DeepStateTrainingNetwork) instance_splitter = self._create_instance_splitter("training") return TrainDataLoader( dataset=data, transform=instance_splitter + SelectFields(input_names), batch_size=self.batch_size, stack_fn=partial(batchify, ctx=self.trainer.ctx, dtype=self.dtype), **kwargs, )
[docs] def create_validation_data_loader( self, data: Dataset, **kwargs, ) -> DataLoader: input_names = get_hybrid_forward_input_names(DeepStateTrainingNetwork) instance_splitter = self._create_instance_splitter("validation") return ValidationDataLoader( dataset=data, transform=instance_splitter + SelectFields(input_names), batch_size=self.batch_size, stack_fn=partial(batchify, ctx=self.trainer.ctx, dtype=self.dtype), )
[docs] def create_training_network(self) -> DeepStateTrainingNetwork: return DeepStateTrainingNetwork( num_layers=self.num_layers, num_cells=self.num_cells, cell_type=self.cell_type, past_length=self.past_length, prediction_length=self.prediction_length, issm=self.issm, dropout_rate=self.dropout_rate, cardinality=self.cardinality, embedding_dimension=self.embedding_dimension, scaling=self.scaling, noise_std_bounds=self.noise_std_bounds, prior_cov_bounds=self.prior_cov_bounds, innovation_bounds=self.innovation_bounds, )
[docs] def create_predictor( self, transformation: Transformation, trained_network: HybridBlock ) -> Predictor: prediction_splitter = self._create_instance_splitter("test") prediction_network = DeepStatePredictionNetwork( num_layers=self.num_layers, num_cells=self.num_cells, cell_type=self.cell_type, past_length=self.past_length, prediction_length=self.prediction_length, issm=self.issm, dropout_rate=self.dropout_rate, cardinality=self.cardinality, embedding_dimension=self.embedding_dimension, scaling=self.scaling, num_parallel_samples=self.num_parallel_samples, noise_std_bounds=self.noise_std_bounds, prior_cov_bounds=self.prior_cov_bounds, innovation_bounds=self.innovation_bounds, params=trained_network.collect_params(), ) copy_parameters(trained_network, prediction_network) return RepresentableBlockPredictor( input_transform=transformation + prediction_splitter, prediction_net=prediction_network, batch_size=self.batch_size, prediction_length=self.prediction_length, ctx=self.trainer.ctx, )