def __init__(self):
super(SimplerCNN, self).__init__()
self.dropout2d_input = nn.Dropout2d(rate=0.3)
self.conv1 = nn.Conv2d(in_channels=3,
out_channels=15,
kernel_size=3,
stride=3,
padding=2)
self.relu1 = nn.LeakyRelu()
self.conv2 = nn.Conv2d(in_channels=15,
out_channels=30,
kernel_size=3,
stride=3,
padding=3)
self.relu2 = nn.LeakyRelu()
self.dropout2d_conv1 = nn.Dropout2d(rate=0.5)
self.conv3 = nn.Conv2d(in_channels=30, out_channels=40, kernel_size=4)
self.relu3 = nn.LeakyRelu()
self.flatten = nn.Flatten()
self.dropout2d_conv2 = nn.Dropout2d(rate=0.2)
self.linear = nn.Linear(in_dimension=360, out_dimension=180)
self.relu4 = nn.LeakyRelu()
self.bn1 = nn.BatchNorm()
self.dropout3 = nn.Dropout(rate=0.3)
self.linear2 = nn.Linear(in_dimension=180, out_dimension=10)
self.bn2 = nn.BatchNorm()
self.softmax = nn.Softmax()
self.set_forward()
def define_model(vocab_size, max_length):
inputs1=m.Input(shape=(4096,))
fe1=m.Dropout(0.5)(inputs1)
fe2=m.Dense(256, activation='relu')(fe1)
inputs2=m.Input(shape=(max_length,))
se1=m.Embedding(vocab_size, 256, mask_zero=True)(inputs2)
se2=m.Dropout(0.5)(se1)
se3=m.LSTM(256)(se2)
decoder1=m.add([fe2, se3])
decoder2=m.Dense(256, activation='relu')(decoder1)
outputs=m.Dense(vocab_size, activation='softmax')(decoder2)
model=m.Model(inputs=[inputs1, inputs2], outputs=outputs)
model.compile(loss='categorical_crossentropy', optimizer='adam')
model.summary()
m.plot_model(model, to_file='model.png', show_shapes=True)
return model
def __init__(self):
super(NN, self).__init__()
self.linear1 = nn.Linear(in_dimension=3072, out_dimension=256)
self.relu1 = nn.LeakyRelu()
self.dropout1 = nn.Dropout(rate=0.3)
self.linear2 = nn.Linear(in_dimension=256, out_dimension=10)
self.softmax = nn.Softmax()
self.set_forward()
def __init__(self, layer_dims, index, position, noise_std, arglist):
super(GRUED, self).__init__()
# this module will hold the variables it needs to in a dictionary
# it will also have a set of functions
self.index = index
self.layer_dims = layer_dims
self.position = position
self.noise_std = noise_std
self.use_bn = True
# encoding modules
if self.position is 'first':
en_indim = self.layer_dims[self.index]
en_outdim = self.layer_dims[self.index]
else:
en_indim = layer_dims[self.index-1]
en_outdim = layer_dims[self.index]
# self.en_conv = nn.Conv2d(en_indim, en_outdim, bias=False, **arglist[self.index-1])
self.en_gru = ConvGRU(en_indim, en_outdim)
self.en_bn_clean = nn.BatchNorm2d(en_outdim, affine=False)
self.en_bn_noisy = nn.BatchNorm2d(en_outdim, affine=False)
self.en_gamma = nn.Parameter(torch.rand(en_outdim, 1, 1))
self.en_beta = nn.Parameter(torch.rand(en_outdim, 1, 1))
self.en_nonlin = nn.Tanh()
# decoding modules
if self.position is 'last':
de_indim = self.layer_dims[self.index]
de_outdim = self.layer_dims[self.index]
else:
de_indim = self.layer_dims[self.index+1]
de_outdim = self.layer_dims[self.index]
self.de_conv = nn.ConvTranspose2d(de_indim, de_outdim, bias=False, **arglist[self.index])
self.de_bn = nn.BatchNorm2d(de_outdim, affine=False)
self.de_gamma = nn.Parameter(torch.rand(de_outdim, 1, 1))
self.de_beta = nn.Parameter(torch.rand(de_outdim, 1, 1))
self.ver_dropout = modules.Dropout(0.5)
self.lat_dropout = modules.Dropout(0.5)
self.parsig1 = modules.ParamSigmoid()
self.parsig2 = modules.ParamSigmoid()