Merge branch 'master' of github.com:awslabs/gluon-ts

src/gluonts/core/serde.py

@@ -476,6 +476,16 @@ def encode_np_dtype(v: np.dtype) -> Any:
  }
 
 
+@encode.register(mx.nd.NDArray)
+def encode_mx_ndarray(v: mx.nd.NDArray) -> Any:
+ return {
+ "__kind__": kind_inst,
+ "class": "mxnet.nd.array",
+ "args": encode([v.asnumpy().tolist()]),
+ "kwargs": {"dtype": encode(v.dtype)},
+ }
+
+
 def decode(r: Any) -> Any:
  """
  Decodes a value from an intermediate representation `r`.

src/gluonts/dataset/artificial/_base.py

@@ -330,7 +330,7 @@ def generate_ts(
  ts_data = dict(
  start=self.start,
  target=target,
- item=str(i),
+ item_id=str(i),
  feat_static_cat=[i],
  feat_static_real=[i],
  )
@@ -508,7 +508,7 @@ def train(self) -> List[DataEntry]:
  dict(
  start=ts["start"],
  target=ts["target"][: -self.prediction_length],
- item=ts["item"],
+ item_id=ts["item_id"],
  )
  for ts in self.make_timeseries()
  ]
@@ -609,7 +609,7 @@ def sigmoid(x: np.ndarray) -> np.ndarray:
  dict(
  start=pd.Timestamp(start, freq=self.freq_str),
  target=np.array(v),
- item=i,
+ item_id=i,
  )
  )
  return res
@@ -681,7 +681,9 @@ def trim_ts_item_end(x: DataEntry, length: int) -> DataEntry:
  the last prediction_length time points from the target and dynamic
  features."""
  y = dict(
- item=x["item"], start=x["start"], target=x["target"][:-length]
+ item_id=x["item_id"],
+ start=x["start"],
+ target=x["target"][:-length],
  )
 
  if "feat_dynamic_cat" in x:
@@ -703,7 +705,7 @@ def trim_ts_item_front(x: DataEntry, length: int) -> DataEntry:
  assert length <= len(x["target"])
 
  y = dict(
- item=x["item"],
+ item_id=x["item_id"],
  start=x["start"] + length * x["start"].freq,
  target=x["target"][length:],
  )
@@ -799,7 +801,7 @@ def constant_dataset() -> Tuple[DatasetInfo, Dataset, Dataset]:
  train_ds = ListDataset(
  data_iter=[
  {
- "item": str(i),
+ "item_id": str(i),
  "start": start_date,
  "target": [float(i)] * 24,
  "feat_static_cat": [i],
@@ -813,7 +815,7 @@ def constant_dataset() -> Tuple[DatasetInfo, Dataset, Dataset]:
  test_ds = ListDataset(
  data_iter=[
  {
- "item": str(i),
+ "item_id": str(i),
  "start": start_date,
  "target": [float(i)] * 30,
  "feat_static_cat": [i],