Merge pull request #997 from jdb78/feature/release

0.10.2 release

Author: jdb78

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Release Notes
 
-## v0.10.2 Multivariate networks (UNRELEASED)
+## v0.10.2 Multivariate networks (23/05/2022)
 
 ### Added
 

docs/source/index.rst

@@ -47,6 +47,12 @@ or to install via conda
 
    conda install pytorch-forecasting pytorch>=1.7 -c pytorch -c conda-forge
 
+To use the MQF2 loss (multivariate quantile loss), also execute
+
+.. code-block::
+
+   pip install git+https://github.com/KelvinKan/CP-Flow.git@package-specific-version --no-deps
+
 Vist :ref:`Getting started <getting-started>` to learn more about the package and detailled installation instruction.
 The :ref:`Tutorials <tutorials>` section provides guidance on how to use models and implement new ones.
 

pytorch_forecasting/models/basic_rnn/__init__.py

@@ -417,23 +417,23 @@ def plot_interpretation(
             plt.Figure: matplotlib figure
         """
         if not isinstance(self.loss, MultiLoss):  # not multi-target
-            prediction = self.to_prediction(dict(prediction=output["prediction"][[idx]].detach().cpu()))[0]
+            prediction = self.to_prediction(dict(prediction=output["prediction"][[idx]].detach()))[0].cpu()
             block_forecasts = [
-                self.to_prediction(dict(prediction=block[[idx]].detach().cpu()))[0]
+                self.to_prediction(dict(prediction=block[[idx]].detach()))[0].cpu()
                 for block in output["block_forecasts"]
             ]
         elif isinstance(output["prediction"], (tuple, list)):  # multi-target
             figs = []
             # predictions and block forecasts need to be converted
-            prediction = [p[[idx]].detach().cpu() for p in output["prediction"]]  # select index
+            prediction = [p[[idx]].detach() for p in output["prediction"]]  # select index
             prediction = self.to_prediction(dict(prediction=prediction))  # transform to prediction
-            prediction = [p[0] for p in prediction]  # select first and only index
+            prediction = [p[0].cpu() for p in prediction]  # select first and only index
 
             block_forecasts = [
-                self.to_prediction(dict(prediction=[b[[idx]].detach().cpu() for b in block]))
+                self.to_prediction(dict(prediction=[b[[idx]].detach() for b in block]))
                 for block in output["block_forecasts"]
             ]
-            block_forecasts = [[b[0] for b in block] for block in block_forecasts]
+            block_forecasts = [[b[0].cpu() for b in block] for block in block_forecasts]
 
             for i in range(len(self.target_names)):
                 if ax is not None: