def run():
start_time = time()
data_cand, data_part, full_data = load_data()
# numeric_parties = full_data.party.map(party_map)
train_c, test_c = train_test_split(data_cand, test_size=0.2)
train_p, test_p = train_test_split(data_part, test_size=0.2)
candidatos_clf = Classifier(train_c.drop('candidatoId', axis=1), train_c.candidatoId)
partidos_clf = Classifier(train_p.drop('idPartido', axis=1), train_p.idPartido)
cand_solver = candidatos_clf._predict()
n_cand, pca_cand_solver = candidatos_clf._pca()
part_solver = partidos_clf._predict()
n_part, pca_part_solver = partidos_clf._pca()
cand_pred = candidatos_clf.classify(test_c.drop('candidatoId', axis=1), test_c.candidatoId, cand_solver)
pca_cand_pred = candidatos_clf.classify(test_c.drop('candidatoId', axis=1), test_c.candidatoId, pca_cand_solver, n_cand)
part_pred = partidos_clf.classify(test_p.drop('idPartido', axis=1), test_p.idPartido, part_solver)
pca_part_pred = partidos_clf.classify(test_p.drop('idPartido', axis=1), test_p.idPartido, pca_part_solver, n_part)
output_results(f'CANDIDATOS | {cand_solver}', test_c.candidatoId, cand_pred)
output_results(f'CANDIDATOS_PCA | {pca_cand_solver}, {n_cand}', test_c.candidatoId, pca_cand_pred)
output_results(f'PARTIDOS | {part_solver}', test_p.idPartido, part_pred)
output_results(f'PARTIDOS_PCA | {pca_part_solver}, {n_part}', test_p.idPartido, pca_part_pred)
cand_part_target, cand_part_pred = candidato_mapper(test_c.candidatoId, cand_pred)
output_results(f'PARTIDOS CON CANDIDATO | {cand_solver}', cand_part_target, cand_part_pred)
cm_cand = ConfusionMatrix(test_c.candidatoId, cand_pred)
cm_pca_cand = ConfusionMatrix(test_c.candidatoId, pca_cand_pred)
cm_part = ConfusionMatrix(test_p.idPartido, part_pred)
cm_pca_part = ConfusionMatrix(test_p.idPartido, pca_part_pred)
cm_cand_part = ConfusionMatrix(cand_part_target, cand_part_pred)
elapsed_time = time() - start_time
print(f'----------------------------------------')
print(f'TOTAL TIME: {datetime.timedelta(seconds=elapsed_time)}')
result = {
'data': {
'candidatos': (train_c, test_c),
'partidos': (train_p, test_p),
},
'results': {
'candidatos': (test_c.candidatoId, cand_pred),
'candidatos_pca': (test_c.candidatoId, pca_cand_pred),
'partidos': (test_p.idPartido, part_pred),
'partidos_pca': (test_p.idPartido, pca_part_pred),
'partidos_candidatos': (cand_part_target, cand_part_pred)
},
'matrices': {
'candidatos': cm_cand,
'candidatos_pca': cm_pca_cand,
'partidos': cm_part,
'partidos_pca': cm_pca_part,
'partidos_candidatos': cm_cand_part
}
}
return result
def train(cfg):
cnn = Classifier(input_shape=cfg["shape"], batch_size=cfg["batch"])
if cfg["model"] == "AlexNet":
cnn.set_default_AlexNet_Model()
else:
arch = cfg["arch"]
cnn.set_custom_model(conv_layers=arch["conv"],
dense_layers=arch["dense"])
if cfg["random"]:
cnn.random_boost = True
if cfg["pre_trained"] is not None:
dir_model = glob(
os.path.join(st.DIR_LOG, "RUNNING",
"*" + cfg["pre_trained"] + "*"))
assert len(dir_model) == 1
# model_file = os.path.join(dir_model[0], "'model-086-1.000000.ckpt")
# cnn.model.load_weights(model_file)
cnn.model.load_weights(os.path.join(dir_model[0], "cp-model.ckpt"))
val_data = cnn.get_data_generator(st.DATA_VALIDATION_DIR,
dip_filter=cfg["dip"])
train_data = cnn.get_data_generator(st.DATA_TRAIN_DIR,
dip_filter=cfg["dip"])
cnn.set_log_name(cfg)
cnn.model_select = cfg["msk"]
cnn.lr["lr"] = cfg["lr"]
cnn.lr["decay_steps"] = cfg["dc_st"]
cnn.lr["decay_rate"] = cfg["dc"]
cnn.lr["momentum"] = cfg["mt"]
cnn.callbacks.append(
TensorBoard(log_dir=os.path.join(st.DIR_LOG, "RUNNING", cnn.log_name)))
cnn.train(gen_train=train_data, gen_val=val_data, epochs=cfg["epochs"])
def run_models(self):
"""run all the models defined in confi file
Save model in pkl format,
Save scores in csv files.
Params:
CONFIG (dict): configurations of task,
SEARCH_SPACE (dict): hyperparameters of models,
df (pandas dataframe): data frame read from raw data,
model: Classifier instance,
all_metrics (list): list to save all model training results.
"""
CONFIG = self.load_config()
DATA_PATH = self.DATA_PATH
SEARCH_SPACE = self.SEARCH_SPACE
df = self.load_data()
model = Classifier(CONFIG, SEARCH_SPACE, df)
all_metrics = []
for model_name in self.CONFIG['RUN_MODELS']:
model_name, grid_cv, best_model, X_test, y_test = model.train_model(
model_name)
metrics = model.gen_metrics(model_name, grid_cv, best_model,
X_test, y_test)
all_metrics.append(metrics)
with open(
f'{self.CONFIG["MODEL_DIR"]}/{model_name}_best_model.pkl',
'wb') as f:
pickle.dump(best_model, f)
df_all_metrics = pd.concat(all_metrics)
df_all_metrics.to_csv(
f'{self.CONFIG["REPORT_DIR"]}/models_metrics.csv', index=False)
def eval_model(cfg):
cnn = Classifier(input_shape=cfg["shape"])
cnn.set_default_AlexNet_Model()
# if cfg["pre_trained"] is not None:
# dir_model = glob(os.path.join(st.DIR_LOG, cfg["log"], "*" + cfg["pre_trained"] + "*"))
# assert len(dir_model) == 1, "the len of chekpoint list is {}".format(len(dir_model))
# cnn.model.load_weights(os.path.join(dir_model[0], "cp-model.ckpt"))
assert cfg["pre_trained"] is not None
dir_model = glob(
os.path.join(st.DIR_LOG, cfg["log"], "*" + cfg["pre_trained"] + "*"))
assert len(dir_model) == 1, "the len of chekpoint list is {}".format(
len(dir_model))
model_file = os.path.join(dir_model[0], "cp-model.ckpt")
# model_file = os.path.join(dir_model[0], "'model-086-1.000000.ckpt")
cnn.model.load_weights(model_file)
dt = Data()
sum_pred = np.zeros(6, )
lenght_data = np.zeros(6, )
for idx in range(dt.length['train']):
image_file, label_gt = dt.get_image_file('train',
idx,
encode_label=True)
label_est = cnn.model_predict(image_file, cfg["dip"])
lenght_data[label_gt] += 1
if label_est == label_gt:
sum_pred[label_gt] += 1
acc = sum_pred / lenght_data
print("{} - Acc: {}".format((label_gt, label_est), acc))
def eval_model(cfg):
cnn = Classifier(input_shape=cfg["shape"])
cnn.set_default_AlexNet_Model()
assert cfg["pre_trained"] is not None
dir_model = glob(
os.path.join(st.DIR_LOG, cfg["log"], "*" + cfg["pre_trained"] + "*"))
assert len(dir_model) == 1, "the len of chekpoint list is {}".format(
len(dir_model))
model_file = os.path.join(dir_model[0], "cp-model.ckpt")
# model_file = os.path.join(dir_model[0], "'model-086-1.000000.ckpt")
cnn.model.load_weights(model_file)
dt = red_csv_file()
filename = os.path.join(
st.DATA_DIR, "eval_test_data_{}_{}.csv".format(cfg["pre_trained"],
cfg["extra"]))
print(filename)
with open(filename, '+w') as csvfile:
writer = csv.writer(csvfile)
writer.writerow(("ID", "Label"))
for idx in range(len(dt)):
image_file = os.path.join(st.DATA_TEST_DIR, dt[idx])
assert os.path.isfile(image_file)
label_est = cnn.model_predict(image_file, cfg["dip"])
with open(filename, 'a') as csvfile:
writer = csv.writer(csvfile)
writer.writerow((dt[idx], label_est))
print("{0:} - progress: {1:.1f}".format(idx, 100 * idx / len(dt)))
def parte_a(train, test, numeric_attributes=IRIS_NUMERIC_ATTRIBUTES):
classifier = Classifier(ID3(train, numeric_attributes))
classifier.train()
actual = []
predicted = []
for _, elem in test.iterrows():
actual.append(elem.clazz)
predicted.append(
classifier.classify(elem.drop(columns=['clazz'], axis=1)))
output_results(title='PARTE A', actual=actual, predicted=predicted)
def run_train():
parser = ArgumentParser()
parser.add_argument('--settings', type=str, required=True,
help='Path to the training settings ini file')
settings = configparser.ConfigParser()
settings.read(parser.parse_args().settings)
# create model
predictor = ResNet50Layers(None)
model = Classifier(predictor)
# use selected gpu by id
gpu = settings.getint('hardware', 'gpu')
if gpu >= 0:
chainer.cuda.get_device_from_id(gpu).use()
model.to_gpu()
label_handler, train_dataset, val_dataset = _create_datasets(settings['input_data'])
train_iter = chainer.iterators.SerialIterator(train_dataset, settings.getint('trainer', 'batchsize'))
val_iter = chainer.iterators.SerialIterator(val_dataset, settings.getint('trainer', 'batchsize'), repeat=False)
output_dir = '{}/training_{}_{}'.format(settings.get('output_data', 'path'), settings.get('trainer', 'epochs'), settings.get('optimizer', 'optimizer'))
# optimizer
optimizer = _create_optimizer(settings['optimizer'])
optimizer.setup(model)
# trainer
updater = chainer.training.updater.StandardUpdater(train_iter, optimizer, device=gpu)
trainer = chainer.training.Trainer(updater, (settings.getint('trainer', 'epochs'), 'epoch'), output_dir)
trainer.extend(extensions.LogReport())
trainer.extend(chainer.training.extensions.ProgressBar(update_interval=1))
evaluator = Evaluator(val_iter, model, device=gpu)
trainer.extend(evaluator)
trainer.extend(extensions.PlotReport(['main/loss', 'validation/main/loss'], x_key='epoch', file_name='loss.png'))
trainer.extend(extensions.PlotReport(['main/accuracy', 'validation/main/accuracy'], x_key='epoch', file_name='accuracy.png'))
trainer.run()
# save model
output_file_path = '{0}/resnet.model'.format(output_dir)
chainer.serializers.save_npz(output_file_path, predictor)
meta_output = {
'trainer': settings._sections['trainer'],
'optimizer': settings._sections['optimizer'],
'train_data': train_dataset.get_meta_info(label_handler),
'validation_data': val_dataset.get_meta_info(label_handler),
}
with open('{0}/meta.json'.format(output_dir), 'w') as f:
json.dump(meta_output, f, indent=4)
def main():
directory = ''
match = False
inject = False
db_name = 'certs.db'
output = False
try:
opts, args = getopt.getopt(sys.argv[1:], 'hd:midb:-o',
['help', 'directory=', 'match', 'inject',
'database=', '--output'])
except:
usage()
for opt, arg in opts:
if opt in ('-h', '--help'):
print('printing hep')
usage()
elif opt in ('-d', '--directory'):
directory = arg
if(not directory.endswith('/')):
directory += '/'
elif opt in ('-m', '--match'):
match = True
elif opt in ('-i', '--inject'):
inject = True
elif opt in ('-b', '--database'):
db = arg
elif opt in ('-o', '--output'):
output = True
outputfile = arg
if(outputfile == ''):
outputfile = 'matches.txt'
if directory:
print("Grabbing data...")
grabber = Grabber(directory)
classifier = Classifier(grabber)
print("Classifying...")
classifier.classify()
certs, keys = classifier.get_data()
db = Database(db_name)
if inject:
try:
print("Creating the database")
db.create_db()
except:
print("Database already exists.")
print("Injecting data into the database...")
db.insert_keys(keys)
db.insert_certs(certs)
if match:
print("Matching data...")
db.match_cert_key()
if output:
db.export_matches(outputfile)
def parte_b(train, test, numeric_attributes=IRIS_NUMERIC_ATTRIBUTES):
classifiers = []
classes = train.clazz.unique()
idx = 1
for clazz in classes:
classifier = Classifier(
ID3(train, numeric_attributes, specific_class=clazz))
classifier.train()
classifiers.append(classifier)
idx += 1
actual = []
predicted = []
for _, elem in test.iterrows():
actual.append(elem.clazz)
predicted.append(
vote_classify(classifiers, elem.drop(columns=['clazz'], axis=1)))
output_results(title='PARTE B', actual=actual, predicted=predicted)
def __init__(self):
# Classifier
self.c = Classifier()
self.c.load_classifier(TRAINING_SET)
self.code = ''
# Correios tracker
self.t = Tracker()
# Answers
with open(GREETINGS_SET) as f:
self.greetings_responses = f.readlines()
with open(COMPLAIN_SET) as f:
self.complain_responses = f.readlines()
with open(COMPLAIN_SET) as f:
self.complain_responses = f.readlines()
with open(HELP_SET) as f:
self.help_responses = f.readlines()
with open(UNKNOWN_SET) as f:
self.unknown_responses = f.readlines()
with open(WRONG_CODE_SET) as f:
self.wrong_code_responses = f.readlines()
with open(TRACKING_SET) as f:
self.tracking_responses = f.readlines()
with open(QUIT_SET) as f:
self.quit_responses = f.readlines()
with open(GOODBYE_SET) as f:
self.goodbye_responses = f.readlines()
with open(RESULTS_SET) as f:
results_responses = f.readlines()
self.offline_response = results_responses[0]
self.fail_response = results_responses[1]
self.sent_response = results_responses[2]
self.forwarded_response = results_responses[3]
self.delivering_response = results_responses[4]
self.arrived_response = results_responses[5]
# If it is expecting something
self.waiting_for = ''
self.attempts = 0
def run(save_loc="cnn/ONN",
loss=Loss.MSE,
OD=10,
gradient=Gradient.APPROXIMATE):
print("\n----- Running {} -----".format(os.path.basename(__file__)))
####################################################
# Configure datasets.
####################################################
dataset = Dataset.MNIST
if dataset != Dataset.MNIST:
save_loc += "_{}".format(str(dataset).split(".")[-1])
batch_size_train = 64
batch_size_test = 1000
####################################################
# Configure Networks.
####################################################
sat_abs_nl_args = {
'I_sat': 1,
'OD': OD,
'encoding': Encoding.AMPLITUDE,
'gradient': gradient
}
SANL = lambda: SatAbsNL(**sat_abs_nl_args)
if loss == Loss.MSE:
output = None
loss_str = "mse"
elif loss == Loss.CCE:
output = lambda: nn.LogSoftmax(-1)
loss_str = "nll"
else:
raise ValueError("Unrecognised loss :", loss)
net_args = {
'n_ch_conv': [32, 64],
'kernel_size_conv': [5, 5],
'n_in_fc': 1024,
'n_hid_fc': [128],
'activation_conv': [SANL, SANL],
'activation_fc': SANL,
'dropout': lambda: nn.Dropout(0.4),
'conv_args': {
'stride': 1,
'padding': 0,
'bias': False
},
'pool_conv': lambda: nn.AvgPool2d(kernel_size=2, stride=2),
'n_out': 10 if dataset != Dataset.EMNIST else 47,
'bias_fc': False,
'output': output
}
####################################################
# Train classifiers
####################################################
n_seeds = 5
losses = {}
corrects = {}
valid_scores = {}
for i in range(n_seeds):
lab = 'seed{}'.format(i)
network = ConvNet(**net_args)
train_loader, test_loader, validation_loader = get_dataset_loaders(
dataset=dataset,
train_batch=batch_size_train,
test_batch=batch_size_test,
unroll_img=False,
max_value=15 if OD > 10 else 5,
get_validation=True)
classifier = Classifier(
network,
train_loader,
test_loader,
n_epochs=30 if dataset == Dataset.MNIST else 40,
learning_rate=5e-4,
init_weight_mean=0.,
init_weight_std=0.01,
init_conv_weight_std=0.1,
loss=loss_str,
weight_range=None,
weight_normalisation=weight_norm.NONE,
log_interval=25,
n_test_per_epoch=0,
save_path=os.path.join(save_loc, lab))
train_losses, test_correct = classifier.train()
losses[lab] = train_losses
corrects[lab] = test_correct
####################################################
# Validation
####################################################
classifier.load(classifier.network_save_path)
valid_loss, valid_correct = classifier.validate(validation_loader)
print("Validation accuracy : {:.2f}%".format(
100. * valid_correct / len(validation_loader.dataset)))
valid_scores[lab] = 100. * valid_correct / len(
validation_loader.dataset)
validation_save_path = os.path.join(classifier.save_path,
"validation_score.pkl")
with open(validation_save_path, 'wb+') as output:
pickle.dump(np.array([valid_loss, valid_correct]), output,
pickle.HIGHEST_PROTOCOL)
print('Validation scores saved to {}'.format(validation_save_path))
print("Validation scores are:")
for lab, score in valid_scores.items():
print("\t{} : {:.2f}%".format(lab, score))
####################################################
# Plot results
####################################################
fig_fname = os.path.join(save_loc, "training_performance")
with plt.style.context('seaborn-paper', after_reset=True):
fig, (ax1, ax2) = plt.subplots(1,
2,
figsize=(7, 2.5),
gridspec_kw={'wspace': 0.3})
window = 25
avg_mask = np.ones(window) / window
for lab, data in losses.items():
ax1.plot(np.convolve(data[:, 0], avg_mask, 'valid'),
np.convolve(data[:, 1], avg_mask, 'valid'),
label=lab,
linewidth=0.75,
alpha=0.8)
ax1.legend()
ax1.set_xlabel("Epoch")
ax1.set_ylabel("Losses")
for lab, data in corrects.items():
ax2.plot(data[:, 0],
data[:, 1] / len(test_loader.dataset),
label=lab)
print("{}: Best score {}/{}".format(lab, np.max(data),
len(test_loader)))
ax2.legend()
ax2.set_xlabel("Epoch")
ax2.set_ylabel("Accuracy")
plt.savefig(fig_fname + ".png", bbox_inches='tight')
plt.savefig(fig_fname + ".pdf", bbox_inches='tight')
def __init__(self, file_name: str):
self.classifier = Classifier(file_name)
self.plot_pair()
def __init__(self, file_name: str, features: list):
self.classifier = Classifier(file_name, filter=False)
self.plotter = Plotter(file_name, features, self.house_array_function)
self.plotter.plot_scatter()
dataset = [x for x in dataset if len(x[0].split()) > 0]
dataset = list(set(dataset))
classifiers = [(SVC(kernel='rbf', C=2.9, gamma=1), 'svm_rbf'),
(SVC(kernel='linear'), 'svm_linear')]
# (KNeighborsClassifier(), 'knn'),
# (MultinomialNB(), 'naive_bayes'),
# (Perceptron(), 'perceptron')]
vectorizers = [(TfidfVectorizer(min_df=0.0, max_df=1.0, sublinear_tf=True, use_idf=True), 'tfidf')]
# (CountVectorizer(min_df=0.0, max_df=1.0), 'count'),
# (HashingVectorizer(), 'hash')]
c = Classifier(classifier=classifiers[1][0], vectorizer=vectorizers[0][0])
x = list(map(lambda a: a[0], dataset))
y = list(map(lambda a: a[1], dataset))
c.train(x_train=x, y_train=y)
from sklearn.pipeline import make_pipeline
import eli5
pipe = make_pipeline(vectorizers[0][0], classifiers[1][0])
pipe.fit(x, y)
file = open('C:/Users/Gustavo/Desktop/batata.html', 'wb')
file.write(eli5.show_weights(classifiers[1][0], vec=vectorizers[0][0], top=1000).data.encode('utf8'))
file.close()
# cv2.waitKey(0)
return image
video_capture = cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
feelings_faces = []
for index, emotion in enumerate(EMOTIONS):
feelings_faces.append(cv2.imread('../emojis/' + emotion + '.png', -1))
while True:
# Capture frame-by-frame
ret, frame = video_capture.read()
# Predict result with network
result = Classifier(format_image(frame)).data[0]
# Draw face in frame
# for (x,y,w,h) in faces:
# cv2.rectangle(frame, (x,y), (x+w,y+h), (255,0,0), 2)
# Write results in frame
if result is not None:
for index, emotion in enumerate(EMOTIONS):
cv2.putText(frame, emotion, (10, index * 20 + 20),
cv2.FONT_HERSHEY_PLAIN, 0.5, (0, 255, 0), 1)
cv2.rectangle(frame, (130, index * 20 + 10),
(130 + int(result[index] * 100),
(index + 1) * 20 + 4), (255, 0, 0), -1)
face_image = feelings_faces[np.argmin(result)]
def run(save_loc="mlp/ANN", n_hid=2, loss=Loss.CCE, activation=nn.ReLU):
print("\n----- Running {} -----".format(os.path.basename(__file__)))
####################################################
# Configure datasets.
####################################################
dataset = Dataset.MNIST
if dataset != Dataset.MNIST:
save_loc += "_{}".format(str(dataset).split(".")[-1])
batch_size_train = 64
batch_size_test = 1000
input_scaling = 1
####################################################
# Configure Networks.
####################################################
if loss == Loss.MSE:
output = None
loss_str = "mse"
elif loss == Loss.CCE:
output = lambda: nn.LogSoftmax(-1)
loss_str = "nll"
else:
raise ValueError("Unrecognised loss :", loss)
default_net_args = {
'n_hid': [128] * n_hid,
'n_in': 784,
'n_out': 10 if dataset != Dataset.EMNIST else 47,
'activation': activation,
'output': output,
}
####################################################
# Train classifiers
####################################################
n_seeds = 5
losses = {}
corrects = {}
valid_scores = {}
for i in range(n_seeds):
lab = 'seed{}'.format(i)
network = LinNet(**default_net_args)
train_loader, test_loader, validation_loader = get_dataset_loaders(
dataset=dataset,
train_batch=batch_size_train,
test_batch=batch_size_test,
unroll_img=True,
max_value=input_scaling,
get_validation=True)
classifier = Classifier(network,
train_loader,
test_loader,
n_epochs=50,
learning_rate=5e-4,
init_weight_mean=0,
init_weight_std=0.1,
loss=loss_str,
weight_range=None,
weight_normalisation=weight_norm.NONE,
log_interval=25,
n_test_per_epoch=0,
save_path=os.path.join(save_loc, lab))
train_losses, test_correct = classifier.train()
losses[lab] = train_losses
corrects[lab] = test_correct
####################################################
# Validation
####################################################
classifier.load(classifier.network_save_path)
valid_loss, valid_correct = classifier.validate(validation_loader)
print("Validation accuracy : {:.2f}%".format(
100. * valid_correct / len(validation_loader.dataset)))
valid_scores[lab] = 100. * valid_correct / len(
validation_loader.dataset)
validation_save_path = os.path.join(classifier.save_path,
"validation_score.pkl")
with open(validation_save_path, 'wb+') as output:
pickle.dump(np.array([valid_loss, valid_correct]), output,
pickle.HIGHEST_PROTOCOL)
print('Validation scores saved to {}'.format(validation_save_path))
print("Validation scores are:")
for lab, score in valid_scores.items():
print("\t{} : {:.2f}%".format(lab, score))
####################################################
# Plot results
####################################################
fig_fname = os.path.join(save_loc, "training_performance")
with plt.style.context('seaborn-paper', after_reset=True):
fig, (ax1, ax2) = plt.subplots(1,
2,
figsize=(7, 2.5),
gridspec_kw={'wspace': 0.3})
window = 25
avg_mask = np.ones(window) / window
for lab, data in losses.items():
ax1.plot(np.convolve(data[:, 0], avg_mask, 'valid'),
np.convolve(data[:, 1], avg_mask, 'valid'),
label=lab,
linewidth=0.75,
alpha=0.8)
ax1.legend()
ax1.set_xlabel("Epoch")
ax1.set_ylabel("Losses")
for lab, data in corrects.items():
ax2.plot(data[:, 0],
data[:, 1] / len(test_loader.dataset),
label=lab)
print("{}: Best score {}/{}".format(lab, np.max(data),
len(test_loader)))
ax2.legend()
ax2.set_xlabel("Epoch")
ax2.set_ylabel("Accuracy")
plt.savefig(fig_fname + ".png", bbox_inches='tight')
plt.savefig(fig_fname + ".pdf", bbox_inches='tight')
def run_train():
parser = ArgumentParser()
parser.add_argument(
'--paths',
type=str,
nargs='+',
required=True,
help='Root paths of folders that contain images and pascal voc files')
parser.add_argument('--label_names',
type=str,
required=True,
help='Path to label names file')
parser.add_argument('--training_splitsize',
type=float,
default=0.9,
help='Splitsize of training data')
parser.add_argument('--batchsize',
type=int,
default=20,
help='Learning minibatch size')
parser.add_argument('--epoch',
type=int,
default=10,
help='Numbers of epochs to train')
parser.add_argument('--gpu',
type=int,
default=-1,
help='GPU ID, negative value indicates CPU')
parser.add_argument('--out',
default='trainer_output',
help='Output directory of trainer')
parser.add_argument('--val_batchsize',
type=int,
default=250,
help='Validation minibatch size')
args = parser.parse_args()
# create model
predictor = ResNet50Layers(None)
model = Classifier(predictor)
# TODO: initmodel
# use selected gpu by id
if args.gpu >= 0:
chainer.cuda.get_device_from_id(args.gpu).use()
model.to_gpu()
# build datasets from paths
label_handler = LabelHandler(args.label_names)
# builder = LabeledImageDatasetBuilder(args.paths, label_handler)
# train_dataset, val_dataset = builder.get_labeled_image_dataset_split(args.training_splitsize)
builder = SortedImageDatasetBuilder(args.paths, label_handler)
train_dataset, val_dataset = builder.get_sorted_image_dataset_split(
args.training_splitsize)
train_iter = chainer.iterators.SerialIterator(train_dataset,
args.batchsize)
val_iter = chainer.iterators.SerialIterator(val_dataset,
args.val_batchsize,
repeat=False)
# optimizer
learning_rate = 0.01
momentum = 0.9
optimizer = chainer.optimizers.MomentumSGD(learning_rate, momentum)
optimizer.setup(model)
# trainer
updater = chainer.training.updater.StandardUpdater(train_iter,
optimizer,
device=args.gpu)
trainer = chainer.training.Trainer(updater, (args.epoch, 'epoch'),
args.out)
trainer.extend(extensions.LogReport())
trainer.extend(chainer.training.extensions.ProgressBar(update_interval=10))
trainer.run()
# save model
output_file_path = '{0}/resnet_{1}_{2}.model'.format(
args.out, args.batchsize, args.epoch)
chainer.serializers.save_npz(output_file_path, predictor)
"""Generate keypoints and descriptions for template images and dump them."""
import cv2 as cv
import os
from os.path import join
from tqdm import tqdm
from src.classifier import Classifier
from src.preprocessing import prepare_image
from util.utils import CLASS_NAMES, FEATURE_DETECTORS, FEATURE_DESCRIPTORS, corner_case
from util.constants import DATA_PATH
if __name__ == '__main__':
template_images = [(name,
prepare_image(cv.imread(join(DATA_PATH,
f"{name}.jpg"))))
for name in CLASS_NAMES]
for detector in FEATURE_DETECTORS:
for descriptor in FEATURE_DESCRIPTORS:
if corner_case(detector, descriptor):
continue
clf = Classifier(detector, descriptor)
for cls_name, image in tqdm(
template_images,
desc=f"Dumping features for {detector}_{descriptor}"):
clf.dump_features(cls_name, image)
from src.stockProcessor import StockProcessor
from src.models import Models
from src.newsProcessor import NewsProcessor
from src.classifier import Classifier
if __name__ == '__main__':
print('load Stock chart data')
base_file_path = sys.argv[1]
stockCharts = StockProcessor(base_file_path)
print('Input stock price interval in minutes - 5, 15, 30, 60, 240, 1440')
time_interval = input()
amazon_stock_prices, apple_stock_prices = stockCharts.loadDataForInterval(time_interval)
For training only
document_vectors_amazon, document_vectors_apple = NewsProcessor(base_file_path).loadNewsArticles()
# AMAZON
classify = Classifier(base_file_path, 'amazon', time_interval, amazon_stock_prices, document_vectors_amazon)
classify.label_documents()
# APPLE
classify = Classifier(base_file_path, 'apple', time_interval, apple_stock_prices, document_vectors_apple)
classify.label_documents()
amazon_model = Models(base_file_path, 'amazon', amazon_stock_prices, time_interval)
amazon_model.naive_bayes_classifier()
amazon_model.SVM_classifier()
amazon_model.DT_classifier()
amazon_model.SVM_poly_classifier()
amazon_model.Logistic_Regression11_classifier()
amazon_model.Logistic_Regression12_classifier()
amazon_model.KNN_classifier()
amazon_model.SGDC_classifier()
amazon_model.accounting_factor()
import os
import cv2 as cv
from os.path import join
from src.classifier import Classifier
from src.inference import draw_bboxes_with_classes
from src.preprocessing import prepare_image, extract_detection, IMG_WIDTH_SIZE
from util.constants import TEST_IMG_PATH, ORB, YOLO_CONFIDENCE, NMS_THRESHOLD
from src.object_detector import Yolo
TEMPLATE_IMGS = ["1uah.jpg", "2uah.jpg", "5uah.jpg", "10uah.jpg"]
if __name__ == '__main__':
for test_image in os.listdir(join(TEST_IMG_PATH, "5uah_heads")):
yolo = Yolo(confidence=YOLO_CONFIDENCE, nms_threshold=NMS_THRESHOLD)
yolo.load_model()
image = cv.imread(join(TEST_IMG_PATH, "5uah_heads", test_image))
prepared_image = prepare_image(image.copy())
img_for_viz = image.copy()
# prepared_image = image
yolo.load_data(prepared_image)
layer_outputs = yolo.detect()
bboxes = yolo.process_outputs(layer_outputs)
clf = Classifier(ORB)
draw_bboxes_with_classes(img_for_viz, bboxes, clf)
cv.imshow("Detections", img_for_viz)
cv.waitKey(0)
cv.destroyAllWindows()
import os
import sys
from src.classifier import Classifier
import tensorflow as tf
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
def print_usage_and_exit():
print("Usage: " + sys.argv[0] +
" <fileClass1> <fileClass2> <corpusDir> <modelFileName>")
sys.exit(0)
if __name__ == '__main__':
if len(sys.argv) != 5:
print_usage_and_exit()
classifier = Classifier(class1_file=sys.argv[1], class2_file=sys.argv[2])
classifier.train(sys.argv[3], sys.argv[4])
from os.path import join
import cv2 as cv
from src.classifier import Classifier
from src.detector_descriptor import DetectorDescriptor
from util.constants import DATA_PATH, TEST_IMG_PATH, ORB
if __name__ == '__main__':
template_img = cv.imread(join(DATA_PATH, "1uah_heads.jpg"))
test_img = cv.imread(join(TEST_IMG_PATH, "1uah_heads", "top.jpeg"))
detector_descriptor = DetectorDescriptor(ORB, ORB)
template_keypoints, template_descriptions = detector_descriptor.detect_describe(
template_img)
test_keypoints, test_descriptions = detector_descriptor.detect_describe(
test_img)
bf_matcher = cv.BFMatcher_create(cv.NORM_HAMMING, True)
matches = bf_matcher.match(template_descriptions, test_descriptions)
clf = Classifier()
print(len(matches))
best_matches = clf.ransac_outlier_rejection(template_keypoints,
test_keypoints, matches)
print(len(best_matches))
return image
if __name__ == '__main__':
parser = _parse_args()
args = parser.parse_args()
if args.image is None and args.video is None:
raise ValueError(
"Neither image nor video arguments are provided! Please,"
" specify either the path to an image or video or both.")
# TODO: supply nms threshold and confidence for YOLO as script arguments
object_detector = Yolo(args.object_detector)
classifier = Classifier(args.feature_detector, args.feature_descriptor,
args.feature_matcher,
args.outlier_rejection_method)
if args.image:
if not isfile(args.image):
raise ValueError(
"The provided path to an image does not exist! "
"Please, provide a path relative to the project:) "
"The directory with test images is test_images/")
image = cv.imread(args.image)
# Image resizing is not necessary,
# OpenCV's YOLO interface takes care of that.
prepared_image = prepare_image(image.copy(), resize=False)
img_for_viz = image.copy()
object_detector.load_data(prepared_image)
# Date: 11/04/2018
# Description: Main file
import os
from src.classifier import Classifier
from werkzeug.utils import secure_filename
from flask import Flask, render_template, request, session, jsonify, redirect, url_for, flash
app = Flask(__name__)
app.config['UPLOAD_FOLDER'] = 'uploads'
app.config['SECRET_KEY'] = os.urandom(0x200)
ALLOWED_EXTENSIONS = set(['png', 'jpg', 'jpeg'])
PATH_TO_MODEL = 'ai/trained_models/cat_dog_3.h5'
classifier = Classifier(PATH_TO_MODEL)
@app.route('/')
def index():
return render_template('index.html')
def allowed_file(filename):
return '.' in filename and \
filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS
@app.route('/upload', methods=['POST'])
def upload():
if 'file' not in request.files:
flash('Upload an image')
set_reproducible()
n_runs = 5
if len(sys.argv) > 1:
n_runs = int(sys.argv[1])
datadir = "../data/"
trainfile = datadir + "traindata.csv"
devfile = datadir + "devdata.csv"
testfile = None
# testfile = datadir + "testdata.csv"
# Runs
start_time = time.perf_counter()
devaccs = []
testaccs = []
for i in range(1, n_runs+1):
classifier = Classifier()
devacc, testacc = train_and_eval(classifier, trainfile, devfile, testfile, i)
devaccs.append(np.round(devacc,2))
testaccs.append(np.round(testacc,2))
print('\nCompleted %d runs.' % n_runs)
total_exec_time = (time.perf_counter() - start_time)
print("Dev accs:", devaccs)
print("Test accs:", testaccs)
print()
print("Mean Dev Acc.: %.2f (%.2f)" % (np.mean(devaccs), np.std(devaccs)))
print("Mean Test Acc.: %.2f (%.2f)" % (np.mean(testaccs), np.std(testaccs)))
print("\nExec time: %.2f s. ( %d per run )" % (total_exec_time, total_exec_time / n_runs))
def clasify(self):
print("step 6. clasifier")
from src.classifier import Classifier
classifier = Classifier()
classifier.parse_arguments(None)
classifier.main()
client = MongoClient('localhost', 27017)
db = client['tcc']
collection = db['tweets']
events = collection.find({"classified": True})
dataset = pickle.load(open(conf.project_path + 'data\dataset_preprocessed.pickle', 'rb'))
dataset = [x for x in dataset if len(x[0].split()) > 0]
# positives = [x for x in dataset if x[1] == 'positive']
# negatives = [x for x in dataset if x[1] == 'negative']
#
# dataset = negatives + positives[0:len(negatives)]
vectorizer = TfidfVectorizer(min_df=0.0, max_df=1.0, sublinear_tf=True, use_idf=True)
classifier = SVC(kernel='rbf', C=2.9, gamma=1)
p = Preprocessor()
clf = Classifier(vectorizer=vectorizer, classifier=classifier)
x_train = [x[0] for x in dataset]
y_train = [x[1] for x in dataset]
clf.train(x_train=x_train, y_train=y_train)
print(clf.predict(p.preprocess('''''')))
# for ev in events:
# print(ev['text'].replace('\n', '') + ' --> ' + clf.predict(p.preprocess(ev['text'])))
