def _build(self):
# TODO Featureless
self.layers.append(
Dense(input_dim=self.dims[0],
output_dim=self.dims[1],
nnz_features=self.data['nnz_features'],
dropout=self.dropouts[0],
act=self.act[0],
bias=self.bias,
sparse_inputs=self.sparse_inputs[0],
logging=self.logging,
model_name=self.name))
for i in range(1, self.n_layers - 1):
self.layers.append(
self.conv_layer(
layer_id=i,
x_names=self.feature_names,
dims=self.dims,
bias=self.bias,
weights=True,
dropout=self.dropouts[i],
# dropout=0.,
act=self.act[i],
shared_weights=self.shared_weights,
nnz_features=self.data['nnz_features'],
sparse_inputs=self.sparse_inputs[i],
skip_connection=self.skip_conn,
add_labels=self.add_labels,
logging=self.logging,
model_name=self.name))
self.layers.append(
Fusion(n_layers=self.n_layers - 2,
x_names=self.feature_names,
input_dim=self.dims[1],
output_dim=self.dims[1],
dropout=self.dropouts[i + 1],
act=tf.nn.relu,
bias=self.bias,
logging=self.logging,
model_name=self.name))
self.layers.append(
Dense(input_dim=self.dims[-2],
output_dim=self.output_dims,
nnz_features=self.data['nnz_features'],
dropout=self.drop_label,
act=lambda x: x,
bias=self.bias,
sparse_inputs=self.sparse_inputs[-1],
logging=self.logging,
model_name=self.name))
def _build(self):
# TODO Featureless
for i in range(self.n_layers):
self.layers.append(
self.conv_layer(layer_id=i,
x_names=self.feature_names,
dims=self.dims,
bias=self.bias,
weights=True,
dropout=self.dropouts[i],
shared_weights=self.shared_weights,
nnz_features=self.data['nnz_features'],
sparse_inputs=self.sparse_inputs[i],
skip_connection=self.skip_conn,
add_labels=self.add_labels,
logging=self.logging,
model_name=self.name))
self.layers.append(
Dense(input_dim=self.dims[-2],
output_dim=self.dims[-1],
nnz_features=None,
dropout=self.dropouts[-1],
act=self.act[-1],
bias=self.bias,
sparse_inputs=self.sparse_inputs[-1],
logging=self.logging))
def main():
input_size = output_size = 2048
now = datetime.datetime.now()
time = now.strftime('%Y.%m.%d %H.%M')
inputs = Input((input_size, ))
layer = Dense(5, activation='relu')(inputs)
layer = Dropout(0.25)(layer)
outputs = Dense(output_size, activation='relu')(layer)
train_x, train_y = make_data(80000, output_size)
test_x, test_y = make_data(20000, output_size)
tensorboard = TensorBoard('./logs/' + time)
model = Model(inputs, outputs)
model.compile('adam', 'mean_squared_error')
model.fit(train_x,
train_y,
epochs=20,
validation_data=(test_x, test_y),
callbacks=[tensorboard])
def _build(self):
# TODO Featureless
self.layers.append(
Dense(input_dim=self.dims[0],
output_dim=self.dims[1],
nnz_features=self.data['nnz_features'],
dropout=self.dropouts[0],
act=self.act[0],
bias=self.bias,
sparse_inputs=self.sparse_inputs[0],
logging=self.logging,
model_name=self.name))
for i in range(1, self.n_layers):
self.layers.append(
Kernel(layer_id=i,
x_names=self.feature_names,
dims=self.dims,
dropout=self.dropouts[i],
act=self.act[i],
bias=self.bias,
shared_weights=self.shared_weights,
nnz_features=self.data['nnz_features'],
sparse_inputs=self.sparse_inputs[i],
skip_connection=self.skip_conn,
add_labels=self.add_labels,
logging=self.logging,
model_name=self.name))
self.layers.append(
Fusion(n_layers=self.n_layers - 1,
x_names=self.feature_names,
input_dim=self.dims[1],
output_dim=self.output_dims,
dropout=self.drop_fuse,
act=(lambda x: x),
bias=self.bias,
logging=self.logging,
model_name=self.name))