[MRG] Fix bug SSW backend (#471)

* fix bug np vs torch matmul

* typo error

* einsum projections ssw

* Test broadcast matmul

* einsum projections ssw

* Test broadcast matmul

* projections SSW with einsum

* reduce number of samples in test wasserstein_circle_unif

* Update releases.md

---------

Co-authored-by: Rémi Flamary <remi.flamary@gmail.com>

Author: clbonet

RELEASES.md

@@ -12,6 +12,7 @@
 - Major documentation cleanup (PR #462, #467)
 - Fix gradients for "Wasserstein2 Minibatch GAN" example (PR #466)
 - Faster Bures-Wasserstein distance with NumPy backend (PR #468)
+- Fix issue backend for ot.sliced_wasserstein_sphere ot.sliced_wasserstein_sphere_unif (PR #471)
 
 ## 0.9.0
 

ot/backend.py

@@ -959,6 +959,14 @@ def detach(self, *args):
         """
         raise NotImplementedError()
 
+    def matmul(self, a, b):
+        r"""
+        Matrix product of two arrays.
+
+        See: https://numpy.org/doc/stable/reference/generated/numpy.matmul.html#numpy.matmul
+        """
+        raise NotImplementedError()
+
 
 class NumpyBackend(Backend):
     """
@@ -1293,6 +1301,9 @@ def detach(self, *args):
             return args[0]
         return args
 
+    def matmul(self, a, b):
+        return np.matmul(a, b)
+
 
 class JaxBackend(Backend):
     """
@@ -1645,6 +1656,9 @@ def detach(self, *args):
             return jax.lax.stop_gradient((args[0],))[0]
         return [jax.lax.stop_gradient((a,))[0] for a in args]
 
+    def matmul(self, a, b):
+        return jnp.matmul(a, b)
+
 
 class TorchBackend(Backend):
     """
@@ -2098,6 +2112,9 @@ def detach(self, *args):
             return args[0].detach()
         return [a.detach() for a in args]
 
+    def matmul(self, a, b):
+        return torch.matmul(a, b)
+
 
 class CupyBackend(Backend):  # pragma: no cover
     """
@@ -2474,6 +2491,9 @@ def detach(self, *args):
             return args[0]
         return args
 
+    def matmul(self, a, b):
+        return cp.matmul(a, b)
+
 
 class TensorflowBackend(Backend):
 
@@ -2865,3 +2885,6 @@ def detach(self, *args):
         if len(args) == 1:
             return tf.stop_gradient(args[0])
         return [tf.stop_gradient(a) for a in args]
+
+    def matmul(self, a, b):
+        return tnp.matmul(a, b)

ot/sliced.py

@@ -260,7 +260,7 @@ def max_sliced_wasserstein_distance(X_s, X_t, a=None, b=None, n_projections=50,
 
 
 def sliced_wasserstein_sphere(X_s, X_t, a=None, b=None, n_projections=50,
-                              p=2, seed=None, log=False):
+                              p=2, projections=None, seed=None, log=False):
     r"""
     Compute the spherical sliced-Wasserstein discrepancy.
 
@@ -287,6 +287,8 @@ def sliced_wasserstein_sphere(X_s, X_t, a=None, b=None, n_projections=50,
         Number of projections used for the Monte-Carlo approximation
     p: float, optional (default=2)
         Power p used for computing the spherical sliced Wasserstein
+    projections: shape (n_projections, dim, 2), optional
+        Projection matrix (n_projections and seed are not used in this case)
     seed: int or RandomState or None, optional
         Seed used for random number generator
     log: bool, optional
@@ -326,22 +328,25 @@ def sliced_wasserstein_sphere(X_s, X_t, a=None, b=None, n_projections=50,
     if nx.any(nx.abs(nx.sum(X_s**2, axis=-1) - 1) > 10**(-4)):
         raise ValueError("X_s is not on the sphere.")
     if nx.any(nx.abs(nx.sum(X_t**2, axis=-1) - 1) > 10**(-4)):
-        raise ValueError("Xt is not on the sphere.")
+        raise ValueError("X_t is not on the sphere.")
 
-    # Uniforms and independent samples on the Stiefel manifold V_{d,2}
-    if isinstance(seed, np.random.RandomState) and str(nx) == 'numpy':
-        Z = seed.randn(n_projections, d, 2)
+    if projections is None:
+        # Uniforms and independent samples on the Stiefel manifold V_{d,2}
+        if isinstance(seed, np.random.RandomState) and str(nx) == 'numpy':
+            Z = seed.randn(n_projections, d, 2)
+        else:
+            if seed is not None:
+                nx.seed(seed)
+            Z = nx.randn(n_projections, d, 2, type_as=X_s)
+
+        projections, _ = nx.qr(Z)
     else:
-        if seed is not None:
-            nx.seed(seed)
-        Z = nx.randn(n_projections, d, 2, type_as=X_s)
-
-    projections, _ = nx.qr(Z)
+        n_projections = projections.shape[0]
 
     # Projection on S^1
     # Projection on plane
-    Xps = nx.transpose(nx.reshape(nx.dot(nx.transpose(projections, (0, 2, 1))[:, None], X_s[:, :, None]), (n_projections, 2, n)), (0, 2, 1))
-    Xpt = nx.transpose(nx.reshape(nx.dot(nx.transpose(projections, (0, 2, 1))[:, None], X_t[:, :, None]), (n_projections, 2, m)), (0, 2, 1))
+    Xps = nx.einsum("ikj, lk -> ilj", projections, X_s)
+    Xpt = nx.einsum("ikj, lk -> ilj", projections, X_t)
 
     # Projection on sphere
     Xps = Xps / nx.sqrt(nx.sum(Xps**2, -1, keepdims=True))
@@ -425,9 +430,11 @@ def sliced_wasserstein_sphere_unif(X_s, a=None, n_projections=50, seed=None, log
 
     # Projection on S^1
     # Projection on plane
-    Xps = nx.transpose(nx.reshape(nx.dot(nx.transpose(projections, (0, 2, 1))[:, None], X_s[:, :, None]), (n_projections, 2, n)), (0, 2, 1))
+    Xps = nx.einsum("ikj, lk -> ilj", projections, X_s)
+
     # Projection on sphere
     Xps = Xps / nx.sqrt(nx.sum(Xps**2, -1, keepdims=True))
+
     # Get coordinates on [0,1[
     Xps_coords = nx.reshape(get_coordinate_circle(nx.reshape(Xps, (-1, 2))), (n_projections, n))
 

test/test_1d_solver.py

@@ -279,7 +279,7 @@ def test_wasserstein1d_circle_devices(nx):
 def test_wasserstein_1d_unif_circle():
     # test semidiscrete_wasserstein2_unif_circle versus wasserstein_circle
     n = 20
-    m = 50000
+    m = 1000
 
     rng = np.random.RandomState(0)
     u = rng.rand(n,)
@@ -298,8 +298,8 @@ def test_wasserstein_1d_unif_circle():
     wass2_unif_circle = ot.semidiscrete_wasserstein2_unif_circle(u, w_u)
 
     # check loss is similar
-    np.testing.assert_allclose(wass2, wass2_unif_circle, atol=1e-3)
-    np.testing.assert_allclose(wass2_circle, wass2_unif_circle, atol=1e-3)
+    np.testing.assert_allclose(wass2, wass2_unif_circle, atol=1e-2)
+    np.testing.assert_allclose(wass2_circle, wass2_unif_circle, atol=1e-2)
 
 
 def test_wasserstein1d_unif_circle_devices(nx):

test/test_backend.py

@@ -298,6 +298,8 @@ def test_empty_backend():
         nx.transpose(M)
     with pytest.raises(NotImplementedError):
         nx.detach(M)
+    with pytest.raises(NotImplementedError):
+        nx.matmul(M, M.T)
 
 
 def test_func_backends(nx):
@@ -308,6 +310,9 @@ def test_func_backends(nx):
     v = rnd.randn(3)
     val = np.array([1.0])
 
+    M1 = rnd.randn(1, 2, 10, 10)
+    M2 = rnd.randn(3, 1, 10, 10)
+
     # Sparse tensors test
     sp_row = np.array([0, 3, 1, 0, 3])
     sp_col = np.array([0, 3, 1, 2, 2])
@@ -326,6 +331,9 @@ def test_func_backends(nx):
         SquareMb = nx.from_numpy(SquareM)
         vb = nx.from_numpy(v)
 
+        M1b = nx.from_numpy(M1)
+        M2b = nx.from_numpy(M2)
+
         val = nx.from_numpy(val)
 
         sp_rowb = nx.from_numpy(sp_row)
@@ -661,6 +669,13 @@ def test_func_backends(nx):
         lst_b.append(nx.to_numpy(B))
         lst_name.append("detach B")
 
+        A = nx.matmul(Mb, Mb.T)
+        lst_b.append(nx.to_numpy(A))
+        lst_name.append("matmul")
+        A = nx.matmul(M1b, M2b)
+        lst_b.append(nx.to_numpy(A))
+        lst_name.append("matmul broadcast")
+
         assert not nx.array_equal(Mb, vb), "array_equal (shape)"
         assert nx.array_equal(Mb, Mb), "array_equal (elements) - expected true"
         assert not nx.array_equal(

test/test_sliced.py

@@ -295,6 +295,26 @@ def test_sliced_sphere_same_dist():
     np.testing.assert_almost_equal(res, 0.)
 
 
+def test_sliced_sphere_same_proj():
+    n_projections = 10
+    n = 100
+    rng = np.random.RandomState(0)
+
+    x = rng.randn(n, 3)
+    x = x / np.sqrt(np.sum(x**2, -1, keepdims=True))
+
+    y = rng.randn(n, 3)
+    y = y / np.sqrt(np.sum(y**2, -1, keepdims=True))
+
+    seed = 42
+
+    cost1, log1 = ot.sliced_wasserstein_sphere(x, y, seed=seed, n_projections=n_projections, log=True)
+    cost2, log2 = ot.sliced_wasserstein_sphere(x, y, seed=seed, n_projections=n_projections, log=True)
+
+    assert np.allclose(log1['projections'], log2['projections'])
+    assert np.isclose(cost1, cost2)
+
+
 def test_sliced_sphere_bad_shapes():
     n = 100
     rng = np.random.RandomState(0)
@@ -398,28 +418,32 @@ def test_sliced_sphere_backend_type_devices(nx):
     y = rng.randn(2 * n, 3)
     y = y / np.sqrt(np.sum(y**2, -1, keepdims=True))
 
+    sw_np, log = ot.sliced_wasserstein_sphere(x, y, log=True)
+    P = log["projections"]
+
     for tp in nx.__type_list__:
         print(nx.dtype_device(tp))
 
         xb, yb = nx.from_numpy(x, y, type_as=tp)
 
-        valb = ot.sliced_wasserstein_sphere(xb, yb)
+        valb = ot.sliced_wasserstein_sphere(xb, yb, projections=nx.from_numpy(P, type_as=tp))
 
         nx.assert_same_dtype_device(xb, valb)
+        np.testing.assert_almost_equal(sw_np, nx.to_numpy(valb))
 
 
 def test_sliced_sphere_gradient():
     if torch:
         import torch.nn.functional as F
 
-        X0 = torch.randn((500, 3))
+        X0 = torch.randn((20, 3))
         X0 = F.normalize(X0, p=2, dim=-1)
         X0.requires_grad_(True)
 
-        X1 = torch.randn((500, 3))
+        X1 = torch.randn((20, 3))
         X1 = F.normalize(X1, p=2, dim=-1)
 
-        sw = ot.sliced_wasserstein_sphere(X1, X0, n_projections=500, p=2)
+        sw = ot.sliced_wasserstein_sphere(X1, X0, n_projections=100, p=2)
         grad_x0 = torch.autograd.grad(sw, X0)[0]
 
         assert not torch.any(torch.isnan(grad_x0))