Register ResNet1d layers in a way compatible with DataParallel

The `self.add_module` is a technique to register parameters, but it does
not work with DataParallel because it does not put the parameters into
the __dict__ in the normal way.
https://discuss.pytorch.org/t/discrepancy-between-manual-parameter-registration-vs-using-nn-modulelist-when-parallelizing/181055
pytorch/pytorch#3174

model.py

@@ -175,14 +175,14 @@ def __init__(self, input_dim, blocks_dim, n_classes, kernel_size=17, dropout_rat
         self.bn1 = nn.BatchNorm1d(n_filters_out)
 
         # Residual block layers
-        self.res_blocks = []
         for i, (n_filters, n_samples) in enumerate(blocks_dim):
             n_filters_in, n_filters_out = n_filters_out, n_filters
             n_samples_in, n_samples_out = n_samples_out, n_samples
             downsample = _downsample(n_samples_in, n_samples_out)
-            resblk1d = ResBlock1d(n_filters_in, n_filters_out, downsample, kernel_size, dropout_rate)
-            self.add_module('resblock1d_{0}'.format(i), resblk1d)
-            self.res_blocks += [resblk1d]
+            setattr(self, f'resblock1d_{i}', ResBlock1d(
+                n_filters_in, n_filters_out, downsample,
+                kernel_size, dropout_rate
+            ))
 
         # Linear layer
         n_filters_last, n_samples_last = blocks_dim[-1]
@@ -198,8 +198,8 @@ def forward(self, x):
 
         # Residual blocks
         y = x
-        for blk in self.res_blocks:
-            x, y = blk(x, y)
+        for i in range(self.n_blk):
+            x, y = getattr(self, f'resblock1d_{i}')(x, y)
 
         # Flatten array
         x = x.view(x.size(0), -1)