Fix example and loss normalization (#87)

* Fix number of features in example

* Fix loss normalization and add test

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

README.md

@@ -30,11 +30,14 @@ for lat in range(-90, 90, 1):
         lat_lons.append((lat, lon))
 model = GraphWeatherForecaster(lat_lons)
 
-features = torch.randn((2, len(lat_lons), 78))
+# Generate 78 random features + 24 non-NWP features (i.e. landsea mask)
+features = torch.randn((2, len(lat_lons), 102))
 
+target = torch.randn((2, len(lat_lons), 78))
 out = model(features)
+
 criterion = NormalizedMSELoss(lat_lons=lat_lons, feature_variance=torch.randn((78,)))
-loss = criterion(out, features)
+loss = criterion(out, target)
 loss.backward()
 ```
 

graph_weather/models/losses.py

@@ -6,7 +6,9 @@
 class NormalizedMSELoss(torch.nn.Module):
     """Loss function described in the paper"""
 
-    def __init__(self, feature_variance: list, lat_lons: list, device="cpu"):
+    def __init__(
+        self, feature_variance: list, lat_lons: list, device="cpu", normalize: bool = False
+    ):
         """
         Normalized MSE Loss as described in the paper
 
@@ -31,6 +33,7 @@ def __init__(self, feature_variance: list, lat_lons: list, device="cpu"):
         for lat, lon in lat_lons:
             weights.append(np.cos(lat * np.pi / 180.0))
         self.weights = torch.tensor(weights, dtype=torch.float)
+        self.normalize = normalize
         assert not torch.isnan(self.weights).any()
 
     def forward(self, pred: torch.Tensor, target: torch.Tensor):
@@ -50,10 +53,11 @@ def forward(self, pred: torch.Tensor, target: torch.Tensor):
         self.feature_variance = self.feature_variance.to(pred.device)
         self.weights = self.weights.to(pred.device)
 
-        # pred = pred / self.feature_variance
-        # target = target / self.feature_variance
-
         out = (pred - target) ** 2
+
+        if self.normalize:
+            out = out / self.feature_variance
+
         assert not torch.isnan(out).any()
         # Mean of the physical variables
         out = out.mean(-1)

tests/test_model.py

@@ -201,3 +201,24 @@ def test_forecaster_and_loss_grad_checkpoint():
     assert not torch.isnan(out).any()
     assert not torch.isnan(out).any()
     loss.backward()
+
+
+def test_normalized_loss():
+    lat_lons = []
+    for lat in range(-90, 90, 5):
+        for lon in range(0, 360, 5):
+            lat_lons.append((lat, lon))
+
+    # Generate output as strictly positive random features
+    out = torch.rand((2, len(lat_lons), 78)) + 0.0001
+    feature_variance = out**2
+    target = torch.zeros((2, len(lat_lons), 78))
+
+    criterion = NormalizedMSELoss(
+        lat_lons=lat_lons, feature_variance=feature_variance, normalize=True
+    )
+    loss = criterion(out, target)
+
+    assert not torch.isnan(loss)
+    # Since feature_variance = out**2 and target = 0, we expect loss = weights
+    assert torch.isclose(loss, criterion.weights.expand_as(out.mean(-1)).mean())