.

harvardnlp · Nov 26, 2019 · a9b4e37 · a9b4e37
1 parent 398ba1a
commit a9b4e37
Show file tree

Hide file tree

Showing 4 changed files with 7 additions and 22 deletions.
diff --git a/torch_struct/linearchain.py b/torch_struct/linearchain.py
@@ -79,7 +79,6 @@ def _dp_scan(self, log_potentials, lengths=None, force_grad=False):
         c.data[:, (~mask).view(-1)] = semiring.zero
         c[:] = semiring.sum(torch.stack([c, lp], dim=-1))
 
-
         # Scan
         for n in range(1, log_N + 1):
             chart = semiring.matmul(chart[:, :, 1::2], chart[:, :, 0::2])

diff --git a/torch_struct/semimarkov.py b/torch_struct/semimarkov.py
@@ -35,8 +35,6 @@ def _dp(self, log_potentials, lengths=None, force_grad=False):
 
         # Init.
         semiring.one_(init.data[:, :, :, 0, 0].diagonal(0, -2, -1))
-        # for k in range(1, K - 1):
-        #     semiring.one_(init[:, :, : , k - 1, k].diagonal(0, -2, -1))
 
         # Length mask
         big = torch.zeros(
@@ -50,33 +48,19 @@ def _dp(self, log_potentials, lengths=None, force_grad=False):
             device=log_potentials.device,
         )
         big[:, :, : N - 1] = log_potentials
-        c = init[:, :, :].view(ssize, batch * bin_N, K -1, K-1, C, C)
+        c = init[:, :, :].view(ssize, batch * bin_N, K - 1, K - 1, C, C)
         lp = big[:, :, :].view(ssize, batch * bin_N, K, C, C)
         mask = torch.arange(bin_N).view(1, bin_N).expand(batch, bin_N)
         mask = mask >= (lengths - 1).view(batch, 1)
         lp.data[:, mask.view(-1)] = semiring.zero
         c.data[:, (~mask).view(-1)] = semiring.zero
-        c[:, :,  :K-1, 0] = semiring.sum(torch.stack([c[:, :,  :K-1, 0],
-                                                      lp[:, :,  1:K]], dim=-1))
-
-        # for b in range(lengths.shape[0]):
-        #     end = lengths[b] - 1
-        #     for k in range(1, K - 1):
-        #         semiring.one_(init[:, b, : end - (k - 1), k - 1, k].diagonal(0, -2, -1))
+        c[:, :, : K - 1, 0] = semiring.sum(
+            torch.stack([c[:, :, : K - 1, 0], lp[:, :, 1:K]], dim=-1)
+        )
 
-        # ks = torch.arange(1, K - 1)
         for k in range(1, K - 1):
             semiring.one_(init[:, :, : (k - 1), k - 1, k].diagonal(0, -2, -1))
 
-        # init[:, :, :end, : (K - 1), 0] = log_potentials[:, :, :end, 1:K]
-
-        # init[:, :, :N-1, : (K - 1), 0] = log_potentials[:, :, :N-1, 1:K]
-        # for b in range(lengths.shape[0]):
-        #     end = lengths[b] - 1
-        #     semiring.one_(init[:, b, end:, 0, 0].diagonal(0, 2, 3))
-        #     init[:, b, :end, : (K - 1), 0] = log_potentials[:, b, :end, 1:K]
-        #     for k in range(1, K - 1):
-        #         semiring.one_(init[:, b, : end - (k - 1), k - 1, k].diagonal(0, 2, 3))
         K_1 = K - 1
 
         # Order n, n-1

diff --git a/torch_struct/semirings/semirings.py b/torch_struct/semirings/semirings.py
@@ -275,6 +275,7 @@ class EntropySemiring(Semiring):
     * Parameter estimation for probabilistic finite-state transducers :cite:`eisner2002parameter`
     * First-and second-order expectation semirings with applications to minimum-risk training on translation forests :cite:`li2009first`
     """
+    zero = 0
 
     @staticmethod
     def size():

diff --git a/torch_struct/test_algorithms.py b/torch_struct/test_algorithms.py
@@ -142,7 +142,8 @@ def test_cky(data):
 @settings(max_examples=50, deadline=None)
 def test_generic_a(data):
     model = data.draw(
-        sampled_from([SemiMarkov]) #Alignment , LinearChain, SemiMarkov, CKY, CKY_CRF, DepTree])
+        sampled_from(
+            [SemiMarkov, Alignment , LinearChain, SemiMarkov, CKY, CKY_CRF, DepTree])
     )
     semiring = data.draw(sampled_from([LogSemiring, MaxSemiring]))
     struct = model(semiring)