fix deepstate serialization, add tests (awslabs#445)

src/gluonts/distribution/lds.py

@@ -28,6 +28,7 @@
 
 
 class ParameterBounds:
+ @validated()
  def __init__(self, lower, upper) -> None:
  assert (
  lower <= upper

src/gluonts/model/deepstate/_estimator.py

@@ -129,7 +129,7 @@ class DeepStateEstimator(GluonEstimator):
  (default: False)
  use_feat_static_cat
  Whether to use the ``feat_static_cat`` field from the data
- (default: False)
+ (default: True)
  embedding_dimension
  Dimension of the embeddings for categorical features
  (default: [min(50, (cat+1)//2) for cat in cardinality])

test/model/test_models.py

@@ -26,6 +26,7 @@
 from gluonts.evaluation.backtest import backtest_metrics
 from gluonts.model.deepar import DeepAREstimator
 from gluonts.model.deep_factor import DeepFactorEstimator
+from gluonts.model.deepstate import DeepStateEstimator
 from gluonts.model.gp_forecaster import GaussianProcessEstimator
 from gluonts.model.npts import NPTSEstimator
 from gluonts.model.predictor import Predictor
@@ -229,6 +230,28 @@ def seasonal_estimator():
  return SeasonalNaiveEstimator, dict(prediction_length=prediction_length)
 
 
+def deepstate_estimator(hybridize: bool = False, batches_per_epoch=1):
+ return (
+ DeepStateEstimator,
+ dict(
+ ctx="cpu",
+ epochs=epochs,
+ learning_rate=1e-2,
+ hybridize=hybridize,
+ num_cells=2,
+ num_layers=1,
+ prediction_length=prediction_length,
+ context_length=2,
+ past_length=prediction_length,
+ cardinality=[1],
+ use_feat_static_cat=False,
+ num_batches_per_epoch=batches_per_epoch,
+ use_symbol_block_predictor=False,
+ num_parallel_samples=2,
+ ),
+ )
+
+
 @flaky(max_runs=3, min_passes=1)
 @pytest.mark.timeout(10) # DeepAR occasionally fails with the 5 second timeout
 @pytest.mark.parametrize(
@@ -252,7 +275,11 @@ def seasonal_estimator():
  + (0.2,),
  ]
  ]
- + [npts_estimator() + (0.0,), seasonal_estimator() + (0.0,)],
+ + [
+ npts_estimator() + (0.0,),
+ seasonal_estimator() + (0.0,),
+ deepstate_estimator(hybridize=False, batches_per_epoch=100) + (0.5,),
+ ],
 )
 def test_accuracy(Estimator, hyperparameters, accuracy):
  estimator = Estimator.from_hyperparameters(freq=freq, **hyperparameters)
@@ -296,6 +323,7 @@ def test_repr(Estimator, hyperparameters):
  mqrnn_estimator(),
  gp_estimator(),
  transformer_estimator(),
+ deepstate_estimator(),
  ],
 )
 def test_serialize(Estimator, hyperparameters):