def invoke(localdir, targetdir, args):
if args.pretrain:
pretrain(args)
if args.train:
fixup(localdir, targetdir, args)
train(args)
if args.pretest:
pretest(args)
if args.test:
test(args)
if args.eval:
evaluation(args)
def main(args):
device = torch.device(
"cuda:0" if torch.cuda.is_available() and args.cuda else "cpu")
results = []
columns = []
for num_quadrant_inputs in args.num_quadrant_inputs:
# adds an s in case of plural quadrants
maybes = "s" if num_quadrant_inputs > 1 else ""
print("Training with {} quadrant{} as input...".format(
num_quadrant_inputs, maybes))
# Dataset
datasets, dataloaders, dataset_sizes = get_data(
num_quadrant_inputs=num_quadrant_inputs, batch_size=128)
# Train baseline
baseline_net = baseline.train(
device=device,
dataloaders=dataloaders,
dataset_sizes=dataset_sizes,
learning_rate=args.learning_rate,
num_epochs=args.num_epochs,
early_stop_patience=args.early_stop_patience,
model_path="baseline_net_q{}.pth".format(num_quadrant_inputs),
)
# Train CVAE
cvae_net = cvae.train(
device=device,
dataloaders=dataloaders,
dataset_sizes=dataset_sizes,
learning_rate=args.learning_rate,
num_epochs=args.num_epochs,
early_stop_patience=args.early_stop_patience,
model_path="cvae_net_q{}.pth".format(num_quadrant_inputs),
pre_trained_baseline_net=baseline_net,
)
# Visualize conditional predictions
visualize(
device=device,
num_quadrant_inputs=num_quadrant_inputs,
pre_trained_baseline=baseline_net,
pre_trained_cvae=cvae_net,
num_images=args.num_images,
num_samples=args.num_samples,
image_path="cvae_plot_q{}.png".format(num_quadrant_inputs),
)
# Retrieve conditional log likelihood
df = generate_table(
device=device,
num_quadrant_inputs=num_quadrant_inputs,
pre_trained_baseline=baseline_net,
pre_trained_cvae=cvae_net,
num_particles=args.num_particles,
col_name="{} quadrant{}".format(num_quadrant_inputs, maybes),
)
results.append(df)
columns.append("{} quadrant{}".format(num_quadrant_inputs, maybes))
results = pd.concat(results, axis=1, ignore_index=True)
results.columns = columns
results.loc["Performance gap", :] = results.iloc[0, :] - results.iloc[1, :]
results.to_csv("results.csv")
def unpickle(file):
import _pickle as cPickle
fo = open(file, 'rb')
dict = cPickle.load(fo, encoding='latin1')
fo.close()
return dict
train_d = unpickle('./cifar-100-python/train')
X = np.reshape(train_d['data'], (50000, 3, 32, 32)).transpose(0, 2, 3, 1)
Y = train_d['fine_labels']
X_train = X[:40000]
Y_train = Y[:40000]
X_validation = X[40000:]
Y_validation = Y[40000:]
Y_train_onehot = []
enc = OneHotEncoder(sparse=False)
# one_hot = [0 for i in range(100)]
Y_train = enc.fit_transform(np.array(Y_train).reshape(-1, 1))
Y_validation = enc.fit_transform(np.array(Y_validation).reshape(-1, 1))
# print Y_train.shape
baseline.train(X_train, Y_train, X_validation, Y_validation)
# import collections
# counter=collections.Counter(Y_train)
# print(counter)
def classification_baseline():
import baseline
enc = OneHotEncoder(sparse=False)
Y_train = enc.fit_transform(np.array(Y_train).reshape(-1, 1))
Y_validation = enc.fit_transform(np.array(Y_validation).reshape(-1, 1))
baseline.train(X_train, Y_train, X_validation, Y_validation)
from datetime import datetime def load_training_data(): with open(commons.AUDIO_BASELINE_TRAIN_X, 'rb') as f: train_x = pickle.load(f) with open(commons.AUDIO_BASELINE_TRAIN_Y, 'rb') as f: train_y = pickle.load(f) with open(commons.AUDIO_BASELINE_TRAIN_WHO, 'r') as f: train_who = [] for line in f.readlines(): train_who.append(line.strip()) return (train_x, train_y, train_who) def load_test_data(): with open(commons.AUDIO_BASELINE_TEST_X, 'rb') as f: test_x = pickle.load(f) with open(commons.AUDIO_BASELINE_TEST_Y, 'rb') as f: test_y = pickle.load(f) with open(commons.AUDIO_BASELINE_TEST_WHO, 'r') as f: test_who = [] for line in f.readlines(): test_who.append(line.strip()) return (test_x, test_y, test_who) if __name__ == '__main__': baseline.train(data=load_training_data(), model_path=commons.AUDIO_BASELINE_MODEL) baseline.test(data=load_test_data(), model_path=commons.AUDIO_BASELINE_MODEL)
train_x.append(x)
train_y.append(y)
elif imdb_id in test_ids:
test_x.append(x)
test_y.append(y)
print("\rDone. ", datetime.now())
with open(FUSION_TRAIN_X, 'wb') as f:
pickle.dump(train_x, f)
with open(FUSION_TRAIN_Y, 'wb') as f:
pickle.dump(train_y, f)
with open(FUSION_TRAIN_WHO, 'w') as f:
for who in train_who:
f.write("{}\n".format(who))
with open(FUSION_TEST_X, 'wb') as f:
pickle.dump(test_x, f)
with open(FUSION_TEST_Y, 'wb') as f:
pickle.dump(test_y, f)
with open(FUSION_TEST_WHO, 'w') as f:
for who in test_who:
f.write("{}\n".format(who))
return (train_x, train_y, train_who), (test_x, test_y, test_who)
if __name__ == '__main__':
train_data, test_data = generate_fusion_data()
baseline.train(data=train_data, model_path=FUSION_MODEL)
baseline.test(data=test_data, model_path=FUSION_MODEL)
def load_training_data():
generate_scene_data()
with open(SCENE_DIR + '/' + TRAIN_X, 'rb') as f:
train_x = pickle.load(f)
with open(SCENE_DIR + '/' + TRAIN_Y, 'rb') as f:
train_y = pickle.load(f)
with open(SCENE_DIR + '/' + TRAIN_WHO, 'r') as f:
train_who = []
for line in f.readlines():
train_who.append(line.strip())
return (train_x, train_y, train_who)
def load_test_data():
generate_scene_data()
with open(SCENE_DIR + '/' + TEST_X, 'rb') as f:
test_x = pickle.load(f)
with open(SCENE_DIR + '/' + TEST_Y, 'rb') as f:
test_y = pickle.load(f)
with open(SCENE_DIR + '/' + TEST_WHO, 'r') as f:
test_who = []
for line in f.readlines():
test_who.append(line.strip())
return (test_x, test_y, test_who)
if __name__ == '__main__':
baseline.train(data=load_training_data(), model_path=SCENE_MODEL)
baseline.test(data=load_test_data(), model_path=SCENE_MODEL)
pickle.dump(test_y, f)
with open(VGGFACE_DIR + '/' + TEST_WHO, 'w') as f:
for who in test_who:
f.write("{}\n".format(who))
def load_training_data():
generate_vggface_data()
with open(VGGFACE_DIR + '/' + TRAIN_X, 'rb') as f:
train_x = pickle.load(f)
with open(VGGFACE_DIR + '/' + TRAIN_Y, 'rb') as f:
train_y = pickle.load(f)
return (train_x, train_y)
def load_test_data():
generate_vggface_data()
with open(VGGFACE_DIR + '/' + TEST_X, 'rb') as f:
test_x = pickle.load(f)
with open(VGGFACE_DIR + '/' + TEST_Y, 'rb') as f:
test_y = pickle.load(f)
with open(VGGFACE_DIR + '/' + TEST_WHO, 'r') as f:
test_who = []
for line in f.readlines():
test_who.append(line.strip())
return (test_x, test_y, test_who)
if __name__ == '__main__':
baseline.train(data=load_training_data(), model_path=VGGFACE_MODEL)
baseline.test(data=load_test_data(), model_path=VGGFACE_MODEL)