learn_conv1d.py

@@ -0,0 +1,31 @@
+# Created on 2018/12/10
+# Author: Kaituo XU
+
+import torch
+torch.manual_seed(123)
+
+Cin, Cout, F, S = 4, 3, 2, 1
+B, Lin = 2, 5
+D = 2
+print('Cin, Cout, F, S, B, Lin, D=', Cin, Cout, F, S, B, Lin, D)
+conv1d = torch.nn.Conv1d(Cin, Cin, F, S, bias=False, padding=(F-1)*D, dilation=D, groups=Cin)
+#conv1d = torch.nn.Conv1d(Cin, Cin, F, S, bias=False, padding=1, dilation=1, groups=Cin)
+inputs = torch.randint(3, (B, Cin, Lin))
+conv1d.weight.data = torch.randint(5, conv1d.weight.size())
+outputs = conv1d(inputs)
+# Lout = (Lin - F) / S + 1
+
+print('weight', conv1d.weight.size())
+print('inputs', inputs.size())
+print('outputs', outputs.size())
+
+print('inputs\n', inputs)
+print('weight\n', conv1d.weight)
+print('outputs\n', outputs)
+print('chomp outputs\n', outputs[:,:,:-(F-1)*D])
+
+# m = torch.nn.Conv1d(16, 33, 3, stride=2)
+# print(m.weight.size())
+# input = torch.randn(20, 16, 50)
+# output = m(input)
+# print(output.size())