def get_params():
global trainCSV
params = request.form.to_dict()
priceSMS = float(params['priceSMS'])
priceEmail = float(params['priceEmail'])
if priceEmail == 0 or priceSMS == 0:
return jsonify({'status': 'Not OK'})
sum_to = float(params['sum'])
withSMS = params['withSMS'] == 'true'
withEmail = params['withEmail'] == 'true'
testCSV = BytesIO(base64.b64decode(params['fileTest']))
model = Model()
if trainCSV == '':
model.load_prev()
else:
trainCSV = BytesIO(base64.b64decode(trainCSV))
model.fit(trainCSV)
model.apply(testCSV)
emails, smss, percents, responses, x = model.optimize(
priceSMS, priceEmail, sum_to, withSMS, withEmail)
return jsonify({
'status': 'OK',
'emails': emails,
'smss': smss,
'percents': percents,
'responses': responses,
'x': x
})
def setup_model(config, embeddings):
""" Sets up the model for training/evaluation. The architecture here extends
on the architecture introduced in the ABCNN paper by allowing for multiple
convolutional layers with different window sizes (computed in parallel, not
in series).
Args:
config: dict
Contains the information needed to setup the model.
embeddings: nn.Embedding
The embedding matrix for the model.
Returns:
model: Model
The instantiated model.
"""
print("Creating the ABCNN model...")
# Create the layers
embeddings_size = config["embeddings"]["size"]
max_length = config["max_length"]
layer_configs = config["layers"]
use_all_layer_outputs = config["use_all_layer_outputs"]
# Initialize the layers
layers = []
layer_sizes = [embeddings_size]
for layer_config in layer_configs:
layer, layer_size = setup_layer(max_length, layer_config)
layers.append(layer)
layer_sizes.append(layer_size)
# Compute the size of the FC layer
final_size = 2 * sum(
layer_sizes) if use_all_layer_outputs else 2 * layer_sizes[-1]
# Put it all together
model = Model(embeddings, layers, use_all_layer_outputs,
final_size).float()
model.apply(weights_init)
return model
