def __init__(self, config, num_labels):
super(BertForSequenceClassification, self).__init__(config)
self.num_labels = num_labels
self.bert = BertModel(config)
self.fc1 = nn.Linear(config.hidden_size, config.hidden_size)
self.fc1_drop = consistent_mc_dropout.ConsistentMCDropout()
self.fc2 = nn.Linear(config.hidden_size, num_labels)
def __init__(self, num_classes):
super().__init__()
self.conv1 = nn.Conv2d(1, 32, kernel_size=5)
self.conv1_drop = consistent_mc_dropout.ConsistentMCDropout2d()
self.conv2 = nn.Conv2d(32, 64, kernel_size=5)
self.conv2_drop = consistent_mc_dropout.ConsistentMCDropout2d()
self.fc1 = nn.Linear(1024, 128)
self.fc1_drop = consistent_mc_dropout.ConsistentMCDropout()
self.fc2 = nn.Linear(128, num_classes)
def __init__(self, embeddings, num_labels):
super(TextCnnMC, self).__init__()
self.char_embedding = nn.Embedding.from_pretrained(embeddings,
freeze=True)
self.convs = nn.ModuleList([
nn.Conv2d(in_channels=1,
out_channels=150,
kernel_size=(k, 150),
padding=(k - 1, 0)) for k in [2, 3, 4]
])
# self.conv_drop = nn.ModuleList(
# [consistent_mc_dropout.ConsistentMCDropout2d(p=0.0) for _ in range(3)]
# )
self.fc1 = nn.Linear(450, 150)
self.fc1_drop = consistent_mc_dropout.ConsistentMCDropout()
self.fc2 = nn.Linear(150, num_labels)
def __init__(self, nc):
super().__init__()
def make_conv(i, o):
return nn.Sequential(
nn.Conv2d(i, o, kernel_size=3, padding=1),
nn.ReLU(),
consistent_mc_dropout.ConsistentMCDropout2d(),
)
self.body = nn.Sequential(
make_conv(1, 32), # 128
make_conv(32, 32),
nn.MaxPool2d(2), # 64
make_conv(32, 32),
make_conv(32, 32),
nn.MaxPool2d(2), # 32
make_conv(32, 16),
make_conv(16, 16),
nn.Flatten(-3, -1),
nn.Linear(16 * 32 * 32, 128),
consistent_mc_dropout.ConsistentMCDropout(),
nn.Linear(128, nc),
nn.LogSoftmax(-1))