From 6d9e8a9a8de62bad3114a3429f7b2de4034459e9 Mon Sep 17 00:00:00 2001
From: jameschapman19 <james.chapman.19@ucl.ac.uk>
Date: Thu, 26 Oct 2023 00:50:45 +0100
Subject: [PATCH] Validation set for Gradient Based Models woohoo

---
 cca_zoo/__init__.py            |  2 ++
 cca_zoo/linear/_tcca.py        | 28 ++++++++++++++++++++--------
 cca_zoo/nonparametric/_ncca.py |  2 +-
 tests/test_unregularized.py    |  3 ++-
 4 files changed, 25 insertions(+), 10 deletions(-)

diff --git a/cca_zoo/__init__.py b/cca_zoo/__init__.py
index 2e7a7678..3928574c 100644
--- a/cca_zoo/__init__.py
+++ b/cca_zoo/__init__.py
@@ -6,6 +6,7 @@
     visualisation,
     preprocessing,
     sequential,
+    nonparametric,
 )
 
 __all__ = [
@@ -16,6 +17,7 @@
     "visualisation",
     "preprocessing",
     "sequential",
+    "nonparametric",
 ]
 try:
     from . import probabilistic
diff --git a/cca_zoo/linear/_tcca.py b/cca_zoo/linear/_tcca.py
index 94a476e2..8f61fd5f 100644
--- a/cca_zoo/linear/_tcca.py
+++ b/cca_zoo/linear/_tcca.py
@@ -96,15 +96,27 @@ def correlations(self, views: Iterable[np.ndarray], **kwargs):
         corrs = multiplied_views / norms
         return corrs
 
-    def score(self, views: Iterable[np.ndarray], **kwargs):
-        """
-        Returns the higher order correlations in each dimension
+    def average_pairwise_correlations(
+        self, views: Iterable[np.ndarray], **kwargs
+    ) -> np.ndarray:
+        transformed_views = self.transform(views, **kwargs)
+        transformed_views = [
+            transformed_view - transformed_view.mean(axis=0)
+            for transformed_view in transformed_views
+        ]
+        multiplied_views = np.stack(transformed_views, axis=0).prod(axis=0).sum(axis=0)
+        norms = np.stack(
+            [
+                np.linalg.norm(transformed_view, axis=0)
+                for transformed_view in transformed_views
+            ],
+            axis=0,
+        ).prod(axis=0)
+        corrs = multiplied_views / norms
+        return corrs
 
-        :param views: list/tuple of numpy arrays or array likes with the same number of rows (samples)
-        :param kwargs: any additional keyword arguments required by the given model
-        """
-        dim_corrs = self.correlations(views, **kwargs)
-        return dim_corrs
+    def score(self, views: Iterable[np.ndarray], **kwargs):
+        return self.average_pairwise_correlations(views, **kwargs).mean()
 
     def _setup_tensor(self, views: Iterable[np.ndarray], **kwargs):
         covs = [
diff --git a/cca_zoo/nonparametric/_ncca.py b/cca_zoo/nonparametric/_ncca.py
index 0b44a43f..c86a89ee 100644
--- a/cca_zoo/nonparametric/_ncca.py
+++ b/cca_zoo/nonparametric/_ncca.py
@@ -5,7 +5,7 @@
 from sklearn.neighbors import NearestNeighbors
 
 from cca_zoo._base import BaseModel
-from cca_zoo.utils.check_values import _process_parameter
+from cca_zoo._utils.check_values import _process_parameter
 
 
 class NCCA(BaseModel):
diff --git a/tests/test_unregularized.py b/tests/test_unregularized.py
index fff09254..25e8da5c 100644
--- a/tests/test_unregularized.py
+++ b/tests/test_unregularized.py
@@ -34,6 +34,8 @@ def test_unregularized_methods(data):
         CCA(latent_dimensions=latent_dims),
         KCCA(latent_dimensions=latent_dims),
         PCACCA(latent_dimensions=latent_dims),
+        TCCA(latent_dimensions=latent_dims),
+        KTCCA(latent_dimensions=latent_dims),
     ]
 
     scores = [
@@ -57,7 +59,6 @@ def test_unregularized_multi(data):
         MCCA(latent_dimensions=latent_dims, pca=False),
         MCCA(latent_dimensions=latent_dims, pca=True),
         KGCCA(latent_dimensions=latent_dims),
-        TCCA(latent_dimensions=latent_dims),
     ]
 
     scores = [method.fit((X, Y, Z)).score((X, Y, Z)) for method in methods]