Merge pull request #32 from NREL/gb/st_expand_bug

Gb/st expand bug

phygnn/layers/custom_layers.py

@@ -132,7 +132,6 @@ def __init__(self, spatial_mult=1):
         """
         super().__init__()
         self._spatial_mult = int(spatial_mult)
-        self._n_spatial_1 = None
 
     @staticmethod
     def _check_shape(input_shape):
@@ -225,9 +224,6 @@ def __init__(self, spatial_mult=1, temporal_mult=1,
         self._spatial_mult = int(spatial_mult)
         self._temporal_mult = int(temporal_mult)
         self._temporal_meth = temporal_method
-        self._n_spatial_1 = None
-        self._n_temporal = None
-        self._temp_expand_shape = None
 
     @staticmethod
     def _check_shape(input_shape):
@@ -254,17 +250,13 @@ def build(self, input_shape):
         """
         self._check_shape(input_shape)
 
-        # desired final shape of the 2nd and 3rd axes for temporal expansion
-        self._n_spatial_1 = input_shape[2]
-        self._n_temporal = input_shape[3]
-        self._temp_expand_shape = tf.stack([
-            self._n_spatial_1, self._n_temporal * self._temporal_mult])
-
     def _temporal_expand(self, x):
         """Expand the temporal dimension (axis=3) of a 5D tensor"""
+        temp_expand_shape = tf.stack(
+            [x.shape[2], x.shape[3] * self._temporal_mult])
         out = []
         for x_unstack in tf.unstack(x, axis=1):
-            out.append(tf.image.resize(x_unstack, self._temp_expand_shape,
+            out.append(tf.image.resize(x_unstack, temp_expand_shape,
                        method=self._temporal_meth))
 
         return tf.stack(out, axis=1)

phygnn/version.py

@@ -1,4 +1,4 @@
 # -*- coding: utf-8 -*-
 """Physics Guided Neural Network version."""
 
-__version__ = '0.0.15'
+__version__ = '0.0.16'

tests/test_base_model.py

@@ -28,7 +28,7 @@ def test_norm_df():
                       label_names=LABELS.columns, normalize=True)
 
     baseline, means, stdevs = PreProcess.normalize(FEATURES)
-    test = model._parse_features(FEATURES)
+    test = model.parse_features(FEATURES)
     assert np.allclose(baseline.values, test)
     assert np.allclose(means, model.feature_means)
     assert np.allclose(stdevs, model.feature_stdevs)
@@ -50,7 +50,7 @@ def test_norm_arr():
                       label_names=label_names, normalize=True)
 
     baseline, means, stdevs = PreProcess.normalize(features)
-    test = model._parse_features(features, names=feature_names)
+    test = model.parse_features(features, names=feature_names)
     assert np.allclose(baseline, test)
     assert np.allclose(means, model.feature_means)
     assert np.allclose(stdevs, model.feature_stdevs)
@@ -77,7 +77,7 @@ def test_OHE():
 
     baseline, means, stdevs = \
         PreProcess.normalize(FEATURES.values.astype('float32'))
-    test = model._parse_features(ohe_features)
+    test = model.parse_features(ohe_features)
 
     assert np.allclose(baseline, test[:, :2])
     assert np.allclose(means,

tests/test_layers.py

@@ -184,8 +184,33 @@ def test_st_expansion(t_mult, s_mult):
     assert y.shape[4] == x.shape[4] / (s_mult**2)
 
 
+def test_st_expansion_new_shape():
+    """Test that the spatiotemporal expansion layer can expand multiple shapes
+    and is not bound to the shape it was built on (bug found on 3/16/2022.)"""
+    s_mult = 3
+    t_mult = 6
+    layer = SpatioTemporalExpansion(spatial_mult=s_mult, temporal_mult=t_mult)
+    n_filters = 2 * s_mult**2
+    x = np.ones((32, 10, 10, 24, n_filters))
+    y = layer(x)
+    assert y.shape[0] == x.shape[0]
+    assert y.shape[1] == s_mult * x.shape[1]
+    assert y.shape[2] == s_mult * x.shape[2]
+    assert y.shape[3] == t_mult * x.shape[3]
+    assert y.shape[4] == x.shape[4] / (s_mult**2)
+
+    x = np.ones((32, 11, 11, 36, n_filters))
+    y = layer(x)
+    assert y.shape[0] == x.shape[0]
+    assert y.shape[1] == s_mult * x.shape[1]
+    assert y.shape[2] == s_mult * x.shape[2]
+    assert y.shape[3] == t_mult * x.shape[3]
+    assert y.shape[4] == x.shape[4] / (s_mult**2)
+
+
 def test_st_expansion_bad():
-    """Test an illegal spatial expansion request."""
+    """Test an illegal spatial expansion request due to number of channels not
+    able to unpack into spatiotemporal dimensions."""
     layer = SpatioTemporalExpansion(spatial_mult=2, temporal_mult=2)
     x = np.ones((123, 10, 10, 24, 3))
     with pytest.raises(RuntimeError):