def get_roc_metrics(data_table: pd.DataFrame, model):
"""Get fpr, tpr and threshold"""
X_train, X_test, y_train, y_test = get_train_and_test_data(data_table)
trained_model = train_model(X_train, y_train, model)
y_prob = get_y_probabilities(X_test, trained_model)
y_pos_prob = y_prob[:, 1]
return roc_curve(y_test, y_pos_prob, pos_label=Values.Y_LABEL_MORE.value)
def test_train_random_forest_model(input_data_features: pd.DataFrame):
"""Tests that random forest model training works"""
X_train, X_test, y_train, y_test = get_train_and_test_data(
input_data_features)
model = Models.RANDOM_FOREST.value
try:
trained_model = train_model(X_train, y_train, model)
except Exception as exception:
pytest.fail('Failed to train model: {}'.format(exception))
open('val_loss_' + args.exp_name + '.csv', 'a') as csv_val, \
open('indv_loss_' + args.exp_name + '.csv', 'a') as csv_indv:
train_writer = csv.writer(csv_train)
test_writer = csv.writer(csv_test)
val_writer = csv.writer(csv_val)
indv_writer = csv.writer(csv_indv)
for epoch in range(args.epochs_start, args.epochs_end):
if args.train:
avg_loss_train = train_model(TileNet,
cuda,
train_dataloader,
optimizer,
epoch + 1,
args.species,
args.alpha,
train_writer,
indv_writer,
margin=margin,
l2=l2,
print_every=print_every,
t0=t0)
train_loss.append(avg_loss_train)
writer.add_scalar('loss/train', avg_loss_train, epoch)
if args.test:
avg_loss_test = validate_model(TileNet,
cuda,
test_dataloader,
optimizer,
epoch + 1,
mse_list = {'big': {}, 'small': {}}
save_dir = paths.model_dir + args.name + '/'
t0 = time()
with open(save_dir + 'command.p', 'wb') as f:
pickle.dump(args, f)
# Train
for epoch in range(args.epochs_start, args.epochs_end):
if args.train:
avg_loss_train = train_model(TileNet,
cuda,
train_dataloader,
optimizer,
epoch + 1,
criterion,
margin=margin,
l2=l2,
print_every=print_every,
t0=t0)
train_loss.append(avg_loss_train)
writer.add_scalar('loss/train', avg_loss_train, epoch)
if args.test:
avg_loss_test = validate_model(TileNet,
cuda,
test_dataloader,
optimizer,
epoch + 1,
criterion,
margin=margin,
parser.add_argument("--wsd_data_dir", required=True, help="Directory where wsd training and"
" evaluation data are located, typically ./wsd_data/")
parser.add_argument("--do_eval", help="If set, the model performs the evaluation step only. "
"Otherwise, it will perform both training and evaluation.",
action="store_true")
parser.add_argument("--epochs", help="Number of training epochs.", type=int, default=40)
args = parser.parse_args()
if args.starting_from_checkpoint:
print("Starting training from epoch {} checkpoint".format(args.starting_epoch))
config = config_class.ConfigAllWords(args.inventory_name, args.model_name, args.starting_epoch, args.data_dir,
args.data_out, args.wsd_data_dir, args.starting_from_checkpoint)
else:
config = config_class.ConfigAllWords(args.inventory_name, args.model_name, args.starting_epoch, args.data_dir,
args.data_out, args.wsd_data_dir)
print('\n\nUsing {} as sense inventory'.format(config.inventory))
print('Output files will be saved to {}'.format(config.experiment_folder))
if not args.do_eval:
utils.define_folders(config)
prepare_data(config)
training.train_model(config, args.epochs)
best_epoch = utils.pick_epoch(config.experiment_folder)
test.test(config, best_epoch)
from src.model_1 import Model1
from src.model_2 import Model2
from src.training import train_model
if __name__ == "__main__":
model1 = Model1()
model2 = Model2()
train_model(model1)
train_model(model2)
from src.training import train_model
if __name__ == "__main__":
train_model()