Allow custom dtype for some VSA similarity functions (#160)

* Allow custom dtype for some VSA similarity functions

* [github-action] formatting fixes

* Increase tolarance for test

* [github-action] formatting fixes

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: mikeheddes

torchhd/functional.py

@@ -893,7 +893,7 @@ def hard_quantize(input: Tensor):
     return torch.where(input > 0, positive, negative)
 
 
-def dot_similarity(input: VSATensor, others: VSATensor) -> VSATensor:
+def dot_similarity(input: VSATensor, others: VSATensor, **kwargs) -> VSATensor:
     """Dot product between the input vector and each vector in others.
 
     Aliased as ``torchhd.dot``.
@@ -938,13 +938,13 @@ def dot_similarity(input: VSATensor, others: VSATensor) -> VSATensor:
     """
     input = ensure_vsa_tensor(input)
     others = ensure_vsa_tensor(others)
-    return input.dot_similarity(others)
+    return input.dot_similarity(others, **kwargs)
 
 
 dot = dot_similarity
 
 
-def cosine_similarity(input: VSATensor, others: VSATensor) -> VSATensor:
+def cosine_similarity(input: VSATensor, others: VSATensor, **kwargs) -> VSATensor:
     """Cosine similarity between the input vector and each vector in others.
 
     Aliased as ``torchhd.cos``.
@@ -987,7 +987,7 @@ def cosine_similarity(input: VSATensor, others: VSATensor) -> VSATensor:
     """
     input = ensure_vsa_tensor(input)
     others = ensure_vsa_tensor(others)
-    return input.cosine_similarity(others)
+    return input.cosine_similarity(others, **kwargs)
 
 
 cos = cosine_similarity

torchhd/tensors/bsbc.py

@@ -334,20 +334,21 @@ def permute(self, shifts: int = 1) -> "BSBCTensor":
         """
         return torch.roll(self, shifts=shifts, dims=-1)
 
-    def dot_similarity(self, others: "BSBCTensor") -> Tensor:
+    def dot_similarity(self, others: "BSBCTensor", *, dtype=None) -> Tensor:
         """Inner product with other hypervectors"""
-        dtype = torch.get_default_dtype()
+        if dtype is None:
+            dtype = torch.get_default_dtype()
 
         if self.dim() > 1 and others.dim() > 1:
             equals = self.unsqueeze(-2) == others.unsqueeze(-3)
             return torch.sum(equals, dim=-1, dtype=dtype)
 
         return torch.sum(self == others, dim=-1, dtype=dtype)
 
-    def cosine_similarity(self, others: "BSBCTensor") -> Tensor:
+    def cosine_similarity(self, others: "BSBCTensor", *, dtype=None) -> Tensor:
         """Cosine similarity with other hypervectors"""
         magnitude = self.size(-1)
-        return self.dot_similarity(others) / magnitude
+        return self.dot_similarity(others, dtype=dtype) / magnitude
 
     @classmethod
     def __torch_function__(cls, func, types, args=(), kwargs=None):

torchhd/tensors/bsc.py

@@ -426,10 +426,11 @@ def permute(self, shifts: int = 1) -> "BSCTensor":
         """
         return super().roll(shifts=shifts, dims=-1)
 
-    def dot_similarity(self, others: "BSCTensor") -> Tensor:
+    def dot_similarity(self, others: "BSCTensor", *, dtype=None) -> Tensor:
         """Inner product with other hypervectors."""
-        dtype = torch.get_default_dtype()
         device = self.device
+        if dtype is None:
+            dtype = torch.get_default_dtype()
 
         min_one = torch.tensor(-1.0, dtype=dtype, device=device)
         plus_one = torch.tensor(1.0, dtype=dtype, device=device)
@@ -441,7 +442,7 @@ def dot_similarity(self, others: "BSCTensor") -> Tensor:
             others_as_bipolar = others_as_bipolar.transpose(-2, -1)
         return torch.matmul(self_as_bipolar, others_as_bipolar)
 
-    def cosine_similarity(self, others: "BSCTensor") -> Tensor:
+    def cosine_similarity(self, others: "BSCTensor", *, dtype=None) -> Tensor:
         """Cosine similarity with other hypervectors."""
         d = self.size(-1)
-        return self.dot_similarity(others) / d
+        return self.dot_similarity(others, dtype=dtype) / d

torchhd/tensors/fhrr.py

@@ -379,6 +379,7 @@ def dot_similarity(self, others: "FHRRTensor") -> Tensor:
         """Inner product with other hypervectors"""
         if others.dim() >= 2:
             others = others.transpose(-2, -1)
+
         return torch.real(torch.matmul(self, torch.conj(others)))
 
     def cosine_similarity(self, others: "FHRRTensor", *, eps=1e-08) -> Tensor:

torchhd/tensors/hrr.py

@@ -366,6 +366,7 @@ def dot_similarity(self, others: "HRRTensor") -> Tensor:
         """Inner product with other hypervectors"""
         if others.dim() >= 2:
             others = others.transpose(-2, -1)
+
         return torch.matmul(self, others)
 
     def cosine_similarity(self, others: "HRRTensor", *, eps=1e-08) -> Tensor:

torchhd/tensors/map.py

@@ -340,16 +340,22 @@ def clipping(self, kappa) -> "MAPTensor":
 
         return torch.clamp(self, min=-kappa, max=kappa)
 
-    def dot_similarity(self, others: "MAPTensor") -> Tensor:
+    def dot_similarity(self, others: "MAPTensor", *, dtype=None) -> Tensor:
         """Inner product with other hypervectors"""
-        dtype = torch.get_default_dtype()
+        if dtype is None:
+            dtype = torch.get_default_dtype()
+
         if others.dim() >= 2:
             others = others.transpose(-2, -1)
+
         return torch.matmul(self.to(dtype), others.to(dtype))
 
-    def cosine_similarity(self, others: "MAPTensor", *, eps=1e-08) -> Tensor:
+    def cosine_similarity(
+        self, others: "MAPTensor", *, dtype=None, eps=1e-08
+    ) -> Tensor:
         """Cosine similarity with other hypervectors"""
-        dtype = torch.get_default_dtype()
+        if dtype is None:
+            dtype = torch.get_default_dtype()
 
         self_dot = torch.sum(self * self, dim=-1, dtype=dtype)
         self_mag = torch.sqrt(self_dot)
@@ -363,4 +369,4 @@ def cosine_similarity(self, others: "MAPTensor", *, eps=1e-08) -> Tensor:
             magnitude = self_mag * others_mag
 
         magnitude = torch.clamp(magnitude, min=eps)
-        return self.dot_similarity(others) / magnitude
+        return self.dot_similarity(others, dtype=dtype) / magnitude

torchhd/tests/basis_hv/test_circular_hv.py

@@ -114,7 +114,9 @@ def test_value(self, dtype, vsa):
 
         elif vsa == "FHRR":
             mag = hv.abs()
-            assert torch.allclose(mag, torch.tensor(1.0, dtype=mag.dtype))
+            assert torch.allclose(
+                mag, torch.tensor(1.0, dtype=mag.dtype), rtol=0.0001, atol=0.0001
+            )
 
         elif vsa == "BSBC":
             assert torch.all((hv >= 0) | (hv < 1024)).item()

torchhd/tests/test_similarities.py

@@ -198,6 +198,37 @@ def test_dtype(self, vsa, dtype):
         else:
             assert similarity.dtype == torch.get_default_dtype()
 
+    def test_custom_dtype(self):
+        hv = functional.random(3, 100, "BSBC", block_size=1024)
+        similarity = functional.dot_similarity(hv, hv)
+        assert similarity.dtype == torch.get_default_dtype()
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.float64)
+        assert similarity.dtype == torch.float64
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.int16)
+        assert similarity.dtype == torch.int16
+
+        hv = functional.random(3, 100, "MAP")
+        similarity = functional.dot_similarity(hv, hv)
+        assert similarity.dtype == torch.get_default_dtype()
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.float64)
+        assert similarity.dtype == torch.float64
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.int16)
+        assert similarity.dtype == torch.int16
+
+        hv = functional.random(3, 100, "BSC")
+        similarity = functional.dot_similarity(hv, hv)
+        assert similarity.dtype == torch.get_default_dtype()
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.float64)
+        assert similarity.dtype == torch.float64
+
+        similarity = functional.dot_similarity(hv, hv, dtype=torch.int16)
+        assert similarity.dtype == torch.int16
+
     @pytest.mark.parametrize("vsa", vsa_tensors)
     @pytest.mark.parametrize("dtype", torch_dtypes)
     def test_device(self, vsa, dtype):