Bump to version 1.7 given new pytorch 1.9 requirement. Stability for …

…DCCA improved!!! Exciting

cca_zoo/deepmodels/dcca.py

@@ -20,7 +20,7 @@ class DCCA(_DCCA_base, torch.nn.Module):
 
     def __init__(self, latent_dims: int, objective=objectives.CCA,
                  encoders: List[BaseEncoder] = [Encoder, Encoder],
-                 learning_rate=1e-3, r: float = 0, eps: float = 1e-9,
+                 learning_rate=1e-3, r: float = 1e-7, eps: float = 1e-7,
                  schedulers: List = None,
                  optimizers: List[torch.optim.Optimizer] = None):
         """

cca_zoo/deepmodels/dccae.py

@@ -20,7 +20,7 @@ class DCCAE(_DCCA_base):
 
     def __init__(self, latent_dims: int, objective=objectives.MCCA,
                  encoders: Iterable[BaseEncoder] = [Encoder, Encoder],
-                 decoders: Iterable[BaseDecoder] = [Decoder, Decoder], r: float = 0, eps: float = 1e-9,
+                 decoders: Iterable[BaseDecoder] = [Decoder, Decoder], r: float = 1e-7, eps: float = 1e-7,
                  learning_rate=1e-3, lam=0.5,
                  schedulers: Iterable = None, optimizers: Iterable = None):
         """

cca_zoo/deepmodels/dtcca.py

@@ -19,7 +19,7 @@ class DTCCA(DCCA, torch.nn.Module):
     """
 
     def __init__(self, latent_dims: int, encoders: Iterable[BaseEncoder] = [Encoder, Encoder],
-                 learning_rate=1e-3, r: float = 0, eps: float = 1e-9,
+                 learning_rate=1e-3, r: float = 1e-7, eps: float = 1e-7,
                  schedulers: Iterable = None, optimizers: Iterable = None):
         """
 

cca_zoo/deepmodels/objectives.py

@@ -21,7 +21,7 @@ class MCCA:
 
     """
 
-    def __init__(self, latent_dims: int, r: float = 0, eps: float = 1e-9):
+    def __init__(self, latent_dims: int, r: float = 1e-7, eps: float = 1e-7):
         """
 
         :param latent_dims: the number of latent dimensions
@@ -73,7 +73,7 @@ class GCCA:
 
     """
 
-    def __init__(self, latent_dims: int, r: float = 0, eps: float = 1e-9):
+    def __init__(self, latent_dims: int, r: float = 1e-7, eps: float = 1e-7):
         """
 
         :param latent_dims: the number of latent dimensions
@@ -123,7 +123,7 @@ class CCA:
 
     """
 
-    def __init__(self, latent_dims: int, r: float = 0, eps: float = 1e-9):
+    def __init__(self, latent_dims: int, r: float = 1e-7, eps: float = 1e-7):
         """
         :param latent_dims: the number of latent dimensions
         :param r: regularisation as in regularized CCA. Makes the problem well posed when batch size is similar to the number of latent dimensions
@@ -173,7 +173,7 @@ class TCCA:
 
     """
 
-    def __init__(self, latent_dims: int, r: float = 0, eps: float = 1e-9):
+    def __init__(self, latent_dims: int, r: float = 1e-7, eps: float = 1e-7):
         """
 
         :param latent_dims: the number of latent dimensions

tests/testdeepmodels.py

@@ -35,6 +35,18 @@ def test_input_types(self):
         dcca_model.fit(self.train_dataset, val_dataset=self.train_dataset, epochs=3)
         dcca_model.fit((self.X, self.Y), val_dataset=(self.X, self.Y), epochs=3)
 
+    def test_large_p(self):
+        X = np.random.rand(2000, 2048)
+        Y = np.random.rand(2000, 2048)
+        latent_dims = 150
+        device = 'cpu'
+        encoder_1 = architectures.Encoder(latent_dims=latent_dims, feature_size=2048)
+        encoder_2 = architectures.Encoder(latent_dims=latent_dims, feature_size=2048)
+        dcca_model = DCCA(latent_dims=latent_dims, encoders=[encoder_1, encoder_2],
+                          objective=objectives.CCA)
+        dcca_model = DeepWrapper(dcca_model, device=device)
+        dcca_model.fit((X, Y), epochs=10)
+
     def test_DCCA_methods_cpu(self):
         latent_dims = 2
         device = 'cpu'