Merge pull request #300 from WenjieDu/dev

Roll back the delta calculation of M-RNN to the same with GRU-D

pypots/imputation/mrnn/data.py

@@ -11,54 +11,7 @@
 from pygrinder import fill_and_get_mask_torch
 
 from ...data.base import BaseDataset
-
-
-def mrnn_parse_delta_torch(missing_mask: torch.Tensor) -> torch.Tensor:
-    """Generate the time-gap matrix from the missing mask, this implementation is the same with the MRNN official
-    implementation in tensorflow https://github.com/jsyoon0823/MRNN, but that is different from the description in the
-    MRNN paper which is the same with the one from GRUD.
-
-    In PyPOTS team's experiments, we find that this implementation is important to the training stability and
-    the performance of MRNN, we think this is mainly because this version make the first step of deltas start from 1,
-    rather than from 0 in the original description.
-
-    Parameters
-    ----------
-    missing_mask : shape of [n_steps, n_features] or [n_samples, n_steps, n_features]
-        Binary masks indicate missing data (0 means missing values, 1 means observed values).
-
-    Returns
-    -------
-    delta :
-        The delta matrix indicates the time gaps between observed values.
-        With the same shape of missing_mask.
-    """
-
-    def cal_delta_for_single_sample(mask: torch.Tensor) -> torch.Tensor:
-        """calculate single sample's delta. The sample's shape is [n_steps, n_features]."""
-        # the first step in the delta matrix is all 0
-        d = [torch.ones(1, n_features, device=device)]
-
-        for step in range(1, n_steps):
-            d.append(
-                torch.ones(1, n_features, device=device) + (1 - mask[step - 1]) * d[-1]
-            )
-        d = torch.concat(d, dim=0)
-        return d
-
-    device = missing_mask.device
-    if len(missing_mask.shape) == 2:
-        n_steps, n_features = missing_mask.shape
-        delta = cal_delta_for_single_sample(missing_mask)
-    else:
-        n_samples, n_steps, n_features = missing_mask.shape
-        delta_collector = []
-        for m_mask in missing_mask:
-            delta = cal_delta_for_single_sample(m_mask)
-            delta_collector.append(delta.unsqueeze(0))
-        delta = torch.concat(delta_collector, dim=0)
-
-    return delta
+from ...data.utils import _parse_delta_torch
 
 
 class DatasetForMRNN(BaseDataset):
@@ -105,10 +58,10 @@ def __init__(
                 forward_missing_mask = self.missing_mask
                 forward_X = self.X
 
-            forward_delta = mrnn_parse_delta_torch(forward_missing_mask)
+            forward_delta = _parse_delta_torch(forward_missing_mask)
             backward_X = torch.flip(forward_X, dims=[1])
             backward_missing_mask = torch.flip(forward_missing_mask, dims=[1])
-            backward_delta = mrnn_parse_delta_torch(backward_missing_mask)
+            backward_delta = _parse_delta_torch(backward_missing_mask)
 
             self.processed_data = {
                 "forward": {
@@ -195,14 +148,14 @@ def _fetch_data_from_file(self, idx: int) -> Iterable:
         forward = {
             "X": X,
             "missing_mask": missing_mask,
-            "deltas": mrnn_parse_delta_torch(missing_mask),
+            "deltas": _parse_delta_torch(missing_mask),
         }
 
         backward = {
             "X": torch.flip(forward["X"], dims=[0]),
             "missing_mask": torch.flip(forward["missing_mask"], dims=[0]),
         }
-        backward["deltas"] = mrnn_parse_delta_torch(backward["missing_mask"])
+        backward["deltas"] = _parse_delta_torch(backward["missing_mask"])
 
         sample = [
             torch.tensor(idx),