Implement GaussProjDrop

egs/librispeech/ASR/pruned_transducer_stateless2/scaling.py

@@ -640,7 +640,7 @@ class ScaledEmbedding(nn.Module):
         return s.format(**self.__dict__)
 
 
-class RandProjDrop(torch.nn.Module):
+class GaussProjDrop(torch.nn.Module):
     """
     This has an effect similar to torch.nn.Dropout, but does not privilege the on-axis directions.
     The directions of dropout are fixed when the class is initialized, and are orthogonal.
@@ -652,8 +652,13 @@ class RandProjDrop(torch.nn.Module):
                  num_channels: int,
                  dropout_rate: float = 0.1,
                  channel_dim: int = -1):
-        super(RandProjDrop, self).__init__()
+        super(GaussProjDrop, self).__init__()
         self.dropout_rate = dropout_rate
+        # this formula for rand_scale was found empirically, trying to match the
+        # statistics of dropout in terms of cross-correlation with the input, see
+        # _test_gauss_proj_drop()
+        self.rand_scale = (dropout_rate / (1-dropout_rate)) ** 0.5 # * (num_channels ** -0.5)
+
         self.channel_dim = channel_dim
 
         rand_mat = torch.randn(num_channels, num_channels)
@@ -661,31 +666,52 @@ class RandProjDrop(torch.nn.Module):
         self.register_buffer('U', U)  # a random orthogonal square matrix.  will be a buffer.
 
 
+    def _randperm_like(self, x: Tensor):
+        """
+        Returns random permutations of the integers [0,1,..x.shape[-1]-1],
+        with the same shape as x.  All dimensions of x other than the last dimension
+        will be treated as batch dimensions.
+
+        Torch's randperm does not support a batch dimension, so we pseudo-randomly simulate it.
+
+        For now, requires x.shape[-1] to be either a power of 2 or 3 times a power of 2, as
+        we normally set channel dims.  This is required for some number theoretic stuff.
+        """
+        n = x.shape[-1]
+
+        assert n & (n-1) == 0 or (n//3 & (n//3 - 1)) == 0
+
+        b = x.numel() // n
+        randint = random.randint(0, 1000)
+        perm = torch.randperm(n, device=x.device)
+        # ensure all elements of batch_rand are coprime to n; this will ensure
+        # that multiplying the permutation by batch_rand and taking modulo
+        # n leaves us with permutations.
+        batch_rand = torch.arange(b, device=x.device) * (randint * 6) + 1
+        batch_rand = batch_rand.unsqueeze(-1)
+        ans = (perm * batch_rand) % n
+        ans = ans.reshape(x.shape)
+        return ans
+
+
     def forward(self, x: Tensor) -> Tensor:
         if not self.training:
-            x = torch.nn.functional.dropout(x, self.dropout_rate,
+            return x
-                                            training=False)
         else:
-            self._randomize_U()
-            x = x.transpose(self.channel_dim, -1)  # (..., num_channels)
-            U = self.U.clone()
-            x = torch.matmul(x, U)
-            x = torch.nn.functional.dropout(x, self.dropout_rate,
-                                            training=True)
-            x = torch.matmul(x, U.t())
             x = x.transpose(self.channel_dim, -1)  # (..., num_channels)
+            x_bypass = x # will be used for "+ I"
+            perm = self._randperm_like(x)
+            x = torch.gather(x, -1, perm)
+            # self.U will act like a different matrix for every row of x, because of the random
+            # permutation.
+            x = torch.matmul(x, self.U)
+            x_next = torch.empty_like(x)
+            # scatter_ uses perm in opposite way
+            # from gather, inverting it.
+            x_next.scatter_(-1, perm, x)
+            x = (x_next * self.rand_scale + x_bypass)
             return x
 
-    def _randomize_U(self):
-        dim = random.randint(0, 1)
-        U = self.U
-        num_channels = U.shape[0]
-        # pick place to split U in two pieces.
-        r = random.randint(1, num_channels - 2)
-        U_part1 = U.narrow(dim, 0, r)
-        U_part2 = U.narrow(dim, r, num_channels-r)
-        U = torch.cat((U_part2, U_part1), dim=dim)
-        self.U[:] = U
 
 def _test_activation_balancer_sign():
     probs = torch.arange(0, 1, 0.01)
@@ -760,22 +786,29 @@ def _test_double_swish_deriv():
     torch.autograd.gradcheck(m, x)
 
 
-def _test_proj_drop():
+def _test_gauss_proj_drop():
-    x = torch.randn(30000, 300)
+    x = torch.randn(30000, 384)
-    m = RandProjDrop(300, 0.1)
+
-    y = m(x)
+
+    for dropout_rate in [0.2, 0.1, 0.01, 0.05]:
+        m1 = torch.nn.Dropout(dropout_rate)
+        m2 = GaussProjDrop(384, dropout_rate)
+        for mode in ['train', 'eval']:
+            y1 = m1(x)
+            y2 = m2(x)
             xmag = (x*x).mean()
-    ymag = (y*y).mean()
+            y1mag = (y1*y1).mean()
-    print(f"xmag = {xmag}, ymag = {ymag}")
+            cross1 = (x*y1).mean()
-    #assert abs((ymag / xmag) - 0.9) < 0.02
+            y2mag = (y2*y2).mean()
-    m.eval()
+            cross2 = (x*y2).mean()
-    y = m(x)
+            print(f"rate={dropout_rate}, mode={mode}, xmag = {xmag}, y1mag = {y1mag}, y2mag = {y2mag}, cross1={cross1}, cross2={cross2}")
-    ymag = (y*y).mean()
+            m1.eval()
-    print(f"xmag[eval] = {xmag}, ymag = {ymag}")
+            m2.eval()
-    #assert abs((ymag / xmag) - 0.9) < 0.02
+
 
 if __name__ == "__main__":
-    _test_proj_drop()
+    _test_gauss_proj_drop()
+    if False:
         _test_activation_balancer_sign()
         _test_activation_balancer_magnitude()
         _test_basic_norm()

egs/librispeech/ASR/pruned_transducer_stateless5/conformer.py

@@ -29,7 +29,7 @@ from scaling import (
     ScaledConv1d,
     ScaledConv2d,
     ScaledLinear,
-    RandProjDrop,
+    GaussProjDrop,
 )
 from torch import Tensor, nn
 
@@ -197,7 +197,7 @@ class ConformerEncoderLayer(nn.Module):
             channel_dim=-1, min_positive=0.45, max_positive=0.55, max_abs=6.0
         )
 
-        self.dropout = RandProjDrop(d_model, dropout)
+        self.dropout = GaussProjDrop(d_model, dropout)
 
     def forward(
         self,
@@ -369,7 +369,7 @@ class RelPositionalEncoding(torch.nn.Module):
         """Construct an PositionalEncoding object."""
         super(RelPositionalEncoding, self).__init__()
         self.d_model = d_model
-        self.dropout = RandProjDrop(d_model, dropout_rate)
+        self.dropout = GaussProjDrop(d_model, dropout_rate)
         self.pe = None
         self.extend_pe(torch.tensor(0.0).expand(1, max_len))