class Inception_Tabular_7_1(nn.Module):
def __init__(self, tabular_model, len_tabular_model_output, initial_kernel_num=64, initialize_parameters=False):
super(Inception_Tabular_7_1, self).__init__()
self.conv_1 = BasicConv2d(in_channels=1, out_channels=64, kernel_size=(7, 1), stride=(2, 1))
self.multi_2d_cnn_1a = nn.Sequential(
Multi_2D_CNN_block(in_channels=64, num_kernel=initial_kernel_num),
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1b = nn.Sequential(
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num * 1.5),
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 1.5),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1c = nn.Sequential(
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 2),
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 2),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2a = nn.Sequential(
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 4),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2b = nn.Sequential(
Multi_2D_CNN_block(in_channels=597, num_kernel=initial_kernel_num * 5),
Multi_2D_CNN_block(in_channels=746, num_kernel=initial_kernel_num * 6),
Multi_2D_CNN_block(in_channels=896, num_kernel=initial_kernel_num * 7),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2c = nn.Sequential(
Multi_2D_CNN_block(in_channels=1045, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2d = nn.Sequential(
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 12),
Multi_2D_CNN_block(in_channels=1792, num_kernel=initial_kernel_num * 14),
Multi_2D_CNN_block(in_channels=2090, num_kernel=initial_kernel_num * 16),
)
self.output_before_fc = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Dropout(0.5),
# nn.Sigmoid()
)
self.tabular_model = tabular_model
self.len_tabular_model_output = len_tabular_model_output
# the extra features are for the output for the tabular data model
self.output = nn.Linear(2389 + len_tabular_model_output, 1)
if initialize_parameters:
for m in self.modules():
model_utils.initialize_parameters(m)
def forward(self, x_and_tabular):
"""
Parameters
----------
x_and_tabular tuple of (image input, tabular features)
where image input and tabular features are both tensors of size BATCHSIZE s.t.
BATCHSIZE is a batch size > 1.
Returns
-------
tensor of BATCHSIZE with forward pass results
"""
x, tabular = x_and_tabular
x = self.conv_1(x)
# N x 1250 x 12 x 64 tensor
x = self.multi_2d_cnn_1a(x)
# N x 416 x 12 x 149 tensor
x = self.multi_2d_cnn_1b(x)
# N x 138 x 12 x 224 tensor
x = self.multi_2d_cnn_1c(x)
# N x 69 x 12 x 298
x = self.multi_2d_cnn_2a(x)
x = self.multi_2d_cnn_2b(x)
x = self.multi_2d_cnn_2c(x)
x = self.multi_2d_cnn_2d(x)
x = self.output_before_fc(x)
tabular_model_output = self.tabular_model(tabular)
len_tabular_model_output = tabular_model_output.shape[1]
assert len_tabular_model_output == self.len_tabular_model_output, \
f"Unexpected length {len_tabular_model_output}"
# see https://discuss.pytorch.org/t/concatenate-layer-output-with-additional-input-data/20462/2
x = torch.cat((x, tabular_model_output), dim=1)
x = self.output(x)
# WARNING: Be careful about .squeeze(). We expect for a given batch, the output is
# [BATCHSIZE, 1] and BATCHSIZE > 1. If the tensor has a batch dimension of size 1,
# then squeeze(input) will also unexpectedly remove the batch dimension!
return x.squeeze()
class Inception_Tabular_15_3(nn.Module):
def __init__(self, tabular_model, len_tabular_model_output, initial_kernel_num=64, initialize_parameters=False):
super(Inception_Tabular_15_3, self).__init__()
self.conv_1 = BasicConv2d(in_channels=1, out_channels=64, kernel_size=(15, 3), stride=(2, 1))
self.multi_2d_cnn_1a = nn.Sequential(
Multi_2D_CNN_block(in_channels=64, num_kernel=initial_kernel_num),
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1b = nn.Sequential(
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num * 1.5),
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 1.5),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1c = nn.Sequential(
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 2),
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 2),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2a = nn.Sequential(
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 4),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2b = nn.Sequential(
Multi_2D_CNN_block(in_channels=597, num_kernel=initial_kernel_num * 5),
Multi_2D_CNN_block(in_channels=746, num_kernel=initial_kernel_num * 6),
Multi_2D_CNN_block(in_channels=896, num_kernel=initial_kernel_num * 7),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2c = nn.Sequential(
Multi_2D_CNN_block(in_channels=1045, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2d = nn.Sequential(
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 12),
Multi_2D_CNN_block(in_channels=1792, num_kernel=initial_kernel_num * 14),
Multi_2D_CNN_block(in_channels=2090, num_kernel=initial_kernel_num * 16),
)
self.output_before_fc = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Dropout(0.5),
# nn.Sigmoid()
)
self.tabular_model = tabular_model
self.len_tabular_model_output = len_tabular_model_output
# the extra features are for the output for the tabular data model
self.output = nn.Linear(2389 + len_tabular_model_output, 1)
if initialize_parameters:
for m in self.modules():
model_utils.initialize_parameters(m)
def forward(self, x_and_tabular):
"""
Parameters
----------
x_and_tabular tuple of (image input, tabular features)
where image input and tabular features are both tensors of size BATCHSIZE s.t.
BATCHSIZE is a batch size > 1.
Returns
-------
tensor of BATCHSIZE with forward pass results
"""
x, tabular = x_and_tabular
x = self.conv_1(x)
# N x 1250 x 12 x 64 tensor
x = self.multi_2d_cnn_1a(x)
# N x 416 x 12 x 149 tensor
x = self.multi_2d_cnn_1b(x)
# N x 138 x 12 x 224 tensor
x = self.multi_2d_cnn_1c(x)
# N x 69 x 12 x 298
x = self.multi_2d_cnn_2a(x)
x = self.multi_2d_cnn_2b(x)
x = self.multi_2d_cnn_2c(x)
x = self.multi_2d_cnn_2d(x)
x = self.output_before_fc(x)
tabular_model_output = self.tabular_model(tabular)
len_tabular_model_output = tabular_model_output.shape[1]
assert len_tabular_model_output == self.len_tabular_model_output, \
f"Unexpected length {len_tabular_model_output}"
# see https://discuss.pytorch.org/t/concatenate-layer-output-with-additional-input-data/20462/2
x = torch.cat((x, tabular_model_output), dim=1)
x = self.output(x)
# WARNING: Be careful about .squeeze(). We expect for a given batch, the output is
# [BATCHSIZE, 1] and BATCHSIZE > 1. If the tensor has a batch dimension of size 1,
# then squeeze(input) will also unexpectedly remove the batch dimension!
return x.squeeze()
class Inception_Tabular_15_1(nn.Module):
def __init__(self, tabular_model, len_tabular_model_output, initial_kernel_num=64, initialize_parameters=False):
super(Inception_Tabular_15_1, self).__init__()
self.conv_1 = BasicConv2d(in_channels=1, out_channels=64, kernel_size=(15, 1), stride=(2, 1))
self.multi_2d_cnn_1a = nn.Sequential(
Multi_2D_CNN_block(in_channels=64, num_kernel=initial_kernel_num),
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1b = nn.Sequential(
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num * 1.5),
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 1.5),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1c = nn.Sequential(
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 2),
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 2),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2a = nn.Sequential(
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 4),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2b = nn.Sequential(
Multi_2D_CNN_block(in_channels=597, num_kernel=initial_kernel_num * 5),
Multi_2D_CNN_block(in_channels=746, num_kernel=initial_kernel_num * 6),
Multi_2D_CNN_block(in_channels=896, num_kernel=initial_kernel_num * 7),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2c = nn.Sequential(
Multi_2D_CNN_block(in_channels=1045, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2d = nn.Sequential(
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 12),
Multi_2D_CNN_block(in_channels=1792, num_kernel=initial_kernel_num * 14),
Multi_2D_CNN_block(in_channels=2090, num_kernel=initial_kernel_num * 16),
)
self.output_before_fc = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Dropout(0.5),
# nn.Sigmoid()
)
self.tabular_model = tabular_model
self.len_tabular_model_output = len_tabular_model_output
# the extra features are for the output for the tabular data model
self.output = nn.Linear(2389 + len_tabular_model_output, 1)
if initialize_parameters:
for m in self.modules():
model_utils.initialize_parameters(m)
def forward(self, x_and_tabular):
"""
Parameters
----------
x_and_tabular tuple of (image input, tabular features)
where image input and tabular features are both tensors of size BATCHSIZE s.t.
BATCHSIZE is a batch size > 1.
Returns
-------
tensor of BATCHSIZE with forward pass results
"""
x, tabular = x_and_tabular
x = self.conv_1(x)
# N x 1250 x 12 x 64 tensor
x = self.multi_2d_cnn_1a(x)
# N x 416 x 12 x 149 tensor
x = self.multi_2d_cnn_1b(x)
# N x 138 x 12 x 224 tensor
x = self.multi_2d_cnn_1c(x)
# N x 69 x 12 x 298
x = self.multi_2d_cnn_2a(x)
x = self.multi_2d_cnn_2b(x)
x = self.multi_2d_cnn_2c(x)
x = self.multi_2d_cnn_2d(x)
x = self.output_before_fc(x)
tabular_model_output = self.tabular_model(tabular)
len_tabular_model_output = tabular_model_output.shape[1]
assert len_tabular_model_output == self.len_tabular_model_output, \
f"Unexpected length {len_tabular_model_output}"
# see https://discuss.pytorch.org/t/concatenate-layer-output-with-additional-input-data/20462/2
x = torch.cat((x, tabular_model_output), dim=1)
x = self.output(x)
# WARNING: Be careful about .squeeze(). We expect for a given batch, the output is
# [BATCHSIZE, 1] and BATCHSIZE > 1. If the tensor has a batch dimension of size 1,
# then squeeze(input) will also unexpectedly remove the batch dimension!
return x.squeeze()
class Inception_Tabular_5_Linears(nn.Module):
def __init__(self, tabular_model, len_tabular_model_output, initial_kernel_num=64, initialize_parameters=False):
self.conv_1 = BasicConv2d(in_channels=1, out_channels=64, kernel_size=(7, 3), stride=(2, 1))
self.multi_2d_cnn_1a = nn.Sequential(
Multi_2D_CNN_block(in_channels=64, num_kernel=initial_kernel_num),
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1b = nn.Sequential(
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num * 1.5),
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 1.5),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1c = nn.Sequential(
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 2),
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 2),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2a = nn.Sequential(
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 4),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2b = nn.Sequential(
Multi_2D_CNN_block(in_channels=597, num_kernel=initial_kernel_num * 5),
Multi_2D_CNN_block(in_channels=746, num_kernel=initial_kernel_num * 6),
Multi_2D_CNN_block(in_channels=896, num_kernel=initial_kernel_num * 7),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2c = nn.Sequential(
Multi_2D_CNN_block(in_channels=1045, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2d = nn.Sequential(
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 12),
Multi_2D_CNN_block(in_channels=1792, num_kernel=initial_kernel_num * 14),
Multi_2D_CNN_block(in_channels=2090, num_kernel=initial_kernel_num * 16),
)
self.output_before_fc = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Dropout(0.5),
# nn.Sigmoid()
)
self.tabular_model = tabular_model
self.len_tabular_model_output = len_tabular_model_output
# the extra features are for the output for the tabular data model
self.output = nn.Sequential(
nn.Linear(2389 + len_tabular_model_output, 1200),
nn.Linear(1200, 400),
nn.Linear(400, 100),
nn.Linear(100, 20),
nn.Linear(20, 1)
),
if initialize_parameters:
for m in self.modules():
model_utils.initialize_parameters(m)
def forward(self, x_and_tabular):
"""
Parameters
----------
x_and_tabular tuple of (image input, tabular features)
where image input and tabular features are both tensors of size BATCHSIZE s.t.
BATCHSIZE is a batch size > 1.
Returns
-------
tensor of BATCHSIZE with forward pass results
"""
x, tabular = x_and_tabular
x = self.conv_1(x)
# N x 1250 x 12 x 64 tensor
x = self.multi_2d_cnn_1a(x)
# N x 416 x 12 x 149 tensor
x = self.multi_2d_cnn_1b(x)
# N x 138 x 12 x 224 tensor
x = self.multi_2d_cnn_1c(x)
# N x 69 x 12 x 298
x = self.multi_2d_cnn_2a(x)
x = self.multi_2d_cnn_2b(x)
x = self.multi_2d_cnn_2c(x)
x = self.multi_2d_cnn_2d(x)
x = self.output_before_fc(x)
tabular_model_output = self.tabular_model(tabular)
len_tabular_model_output = tabular_model_output.shape[1]
assert len_tabular_model_output == self.len_tabular_model_output, \
f"Unexpected length {len_tabular_model_output}"
# see https://discuss.pytorch.org/t/concatenate-layer-output-with-additional-input-data/20462/2
x = torch.cat((x, tabular_model_output), dim=1)
x = self.output(x)
# WARNING: Be careful about .squeeze(). We expect for a given batch, the output is
# [BATCHSIZE, 1] and BATCHSIZE > 1. If the tensor has a batch dimension of size 1,
# then squeeze(input) will also unexpectedly remove the batch dimension!
return x.squeeze()
class Inception_Tabular_5_Linears_BatchNorm(nn.Module):
def __init__(self, tabular_model, len_tabular_model_output, initial_kernel_num=64, initialize_parameters=False):
self.conv_1 = BasicConv2d(in_channels=1, out_channels=64, kernel_size=(7, 3), stride=(2, 1))
self.multi_2d_cnn_1a = nn.Sequential(
Multi_2D_CNN_block(in_channels=64, num_kernel=initial_kernel_num),
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1b = nn.Sequential(
Multi_2D_CNN_block(in_channels=149, num_kernel=initial_kernel_num * 1.5),
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 1.5),
nn.MaxPool2d(kernel_size=(3, 1))
)
self.multi_2d_cnn_1c = nn.Sequential(
Multi_2D_CNN_block(in_channels=224, num_kernel=initial_kernel_num * 2),
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 2),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2a = nn.Sequential(
Multi_2D_CNN_block(in_channels=298, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 3),
Multi_2D_CNN_block(in_channels=448, num_kernel=initial_kernel_num * 4),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2b = nn.Sequential(
Multi_2D_CNN_block(in_channels=597, num_kernel=initial_kernel_num * 5),
Multi_2D_CNN_block(in_channels=746, num_kernel=initial_kernel_num * 6),
Multi_2D_CNN_block(in_channels=896, num_kernel=initial_kernel_num * 7),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2c = nn.Sequential(
Multi_2D_CNN_block(in_channels=1045, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 8),
nn.MaxPool2d(kernel_size=(2, 1))
)
self.multi_2d_cnn_2d = nn.Sequential(
Multi_2D_CNN_block(in_channels=1194, num_kernel=initial_kernel_num * 12),
Multi_2D_CNN_block(in_channels=1792, num_kernel=initial_kernel_num * 14),
Multi_2D_CNN_block(in_channels=2090, num_kernel=initial_kernel_num * 16),
)
self.output_before_fc = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Dropout(0.5),
# nn.Sigmoid()
)
self.tabular_model = tabular_model
self.len_tabular_model_output = len_tabular_model_output
# the extra features are for the output for the tabular data model
self.output = nn.Sequential(
nn.Linear(2389 + len_tabular_model_output, 1200),
nn.BatchNorm1d(1200),
nn.Linear(1200, 400),
nn.BatchNorm1d(400),
nn.Linear(400, 100),
nn.BatchNorm1d(100),
nn.Linear(100, 20),
nn.BatchNorm1d(20),
nn.Linear(20, 1)
),
if initialize_parameters:
for m in self.modules():
model_utils.initialize_parameters(m)
def forward(self, x_and_tabular):
"""
Parameters
----------
x_and_tabular tuple of (image input, tabular features)
where image input and tabular features are both tensors of size BATCHSIZE s.t.
BATCHSIZE is a batch size > 1.
Returns
-------
tensor of BATCHSIZE with forward pass results
"""
x, tabular = x_and_tabular
x = self.conv_1(x)
# N x 1250 x 12 x 64 tensor
x = self.multi_2d_cnn_1a(x)
# N x 416 x 12 x 149 tensor
x = self.multi_2d_cnn_1b(x)
# N x 138 x 12 x 224 tensor
x = self.multi_2d_cnn_1c(x)
# N x 69 x 12 x 298
x = self.multi_2d_cnn_2a(x)
x = self.multi_2d_cnn_2b(x)
x = self.multi_2d_cnn_2c(x)
x = self.multi_2d_cnn_2d(x)
x = self.output_before_fc(x)
tabular_model_output = self.tabular_model(tabular)
len_tabular_model_output = tabular_model_output.shape[1]
assert len_tabular_model_output == self.len_tabular_model_output, \
f"Unexpected length {len_tabular_model_output}"
# see https://discuss.pytorch.org/t/concatenate-layer-output-with-additional-input-data/20462/2
x = torch.cat((x, tabular_model_output), dim=1)
x = self.output(x)
# WARNING: Be careful about .squeeze(). We expect for a given batch, the output is
# [BATCHSIZE, 1] and BATCHSIZE > 1. If the tensor has a batch dimension of size 1,
# then squeeze(input) will also unexpectedly remove the batch dimension!
return x.squeeze()
Datasets
Models
