def train(self, epochs, verbose=False):
self.model.train()
optimizer = torch.optim.AdamW([
{
'params': self.model.resnet50.parameters(),
'lr': 0.001
}, # Low learning rate for the pretrained layers
{
'params': self.model.fc.parameters(),
'lr': 0.01
} # Higher learning rate for the final classifying layers
])
epoch_losses = []
AUCs = []
for epoch in range(1, epochs + 1):
epoch_loss = 0
epoch_start_time = time.time()
batch_start_time = time.time()
epoch_samples_count = 0
for step, (X, y) in enumerate(
self.train_loader): # for each training step
X, y = X.float(), y.float(
) # Convert as they're stored as doubles
X, y = Variable(X).to(self.device), Variable(y).to(
self.device) # Bring to GPU
prediction = self.model(X) # input x and predict based on x
loss = self.criterion(prediction, y) # compute loss
epoch_loss += loss.item() * prediction.shape[0]
optimizer.zero_grad() # clear gradients for next train
loss.backward() # backpropagate
optimizer.step() # apply gradients
batch_duration = time.time() - batch_start_time
epoch_duration = time.time() - epoch_start_time
batch_start_time = time.time()
epoch_samples_count += len(X)
sys.stdout.write(
f"\rEpoch {epoch} - ({epoch_samples_count}/{len(self.train_loader.dataset)}) - Loss: {loss.item():.3f} - epoch: {epoch_duration:.3f}s - step: {batch_duration:.3f}s"
) # Cute inline logging
if step % 10 == 0 and verbose:
print(
f"Epoch {epoch} - ({epoch_samples_count}/{len(self.train_loader.dataset)}) - Last prediction: {prediction} vs {y}"
)
epoch_time = time.time() - epoch_start_time
epoch_losses.append(epoch_loss)
print(
f"\nEpoch {epoch} done. Average loss: {(epoch_loss/len(self.train_loader.dataset)):.3f} - {epoch_time:.4f}s"
)
if verbose:
print("Last prediction", prediction)
print("Last y", y)
auc = aggregate.evaluate(
self.aggregator, self.model
) # Evaluate results using the AUC of the ROC for the model on the test data
AUCs.append(auc)
models.save_model(self.model, self.models_path, epoch)
models.save_model(self.model, self.models_path, epoch)
print("Training complete.")
print(f"Areas under ROC Curves for each epoch: \n{AUCs}")
return epoch_losses
def train_model(dataset, transform_type):
"""
Train specific model on given dataset.
:param dataset:
:param transform_type:
"""
print('Training model ({}) on {}...'.format(transform_type, dataset))
(X_train, Y_train), (X_test, Y_test) = load_data(dataset)
nb_examples, img_rows, img_cols, nb_channels = X_train.shape
nb_classes = Y_train.shape[1]
input_shape = (img_rows, img_cols, nb_channels)
X_train = transform(X_train, transform_type)
model_name = 'model-{}-cnn-{}'.format(dataset, transform_type)
require_preprocess = False
if (dataset == DATA.cifar_10):
require_preprocess = True
# train
model = models.create_model(dataset, input_shape, nb_classes)
models.train(model, X_train, Y_train, model_name, require_preprocess)
# save to disk
models.save_model(model, model_name)
# evaluate the new model
X_test = transform(X_test, transform_type)
loaded_model = models.load_model(model_name)
scores = loaded_model.evaluate(X_test, Y_test, verbose=2)
print('*** Evaluating the new model: {}'.format(scores))
del loaded_model
def main_train(model, m0, vocab, seqs, n_epochs=1, l_rate=1 / 10**6):
split_i = 1
avg_acc = 0
avg_acc3 = 0
start_time = get_time()
n_splits = len(seqs)
model_name = m0["model_name"]
b_size = m0["batch_size"]
word2id = vocab["word2id"]
for seqs_train, seqs_val in seqs:
print("\n")
print("********train split[" + str(split_i) + "/" + str(n_splits) +
"]")
print("train_split length : " + str(len(seqs_train[0])))
print("val_split length : " + str(len(seqs_val[0])))
t_data = n_epochs, seqs_train, seqs_val
model, acc, acc3 = train_model(t_data, model, l_rate, b_size, word2id,
device)
avg_acc += acc
avg_acc3 += acc3
split_i += 1
avg_acc = avg_acc / n_splits
avg_acc3 = avg_acc3 / n_splits
print("")
print("**avg accuracy : " + str(round(100 * avg_acc, 2)) + "%")
print("**avg accuracy3 : " + str(round(100 * avg_acc3, 2)) + "%")
print("**total time taken : " + comp_time(start_time, None))
if model_name != None:
save_model(model_name, model)
save_acc(model_name, n_epochs, avg_acc, avg_acc3)
def train(config):
# train_path:train-context.json
args = config.args
train_set = get_dataset(config.train_path,
config.w2i_vocabs,
config,
is_train=True)
dev_set = get_dataset(config.dev_path,
config.w2i_vocabs,
config,
is_train=False)
# X:img,torch.stack;
train_batch = get_dataloader(train_set, args.batch_size, is_train=True)
model = Model(n_emb=args.n_emb,
n_hidden=args.n_hidden,
vocab_size=args.vocab_size,
dropout=args.dropout,
d_ff=args.d_ff,
n_head=args.n_head,
n_block=args.n_block)
if args.restore != '':
model_dict = torch.load(args.restore)
model.load_state_dict(model_dict)
model.to(device)
optimizer = optim.Adam(filter(lambda p: p.requires_grad,
model.parameters()),
lr=args.lr)
best_score = -1000000
for i in range(args.epoch):
model.train()
report_loss, start_time, n_samples = 0, time.time(), 0
count, total = 0, len(train_set) // args.batch_size + 1
for batch in train_batch:
Y, T = batch
Y = Y.to(device)
T = T.to(device)
optimizer.zero_grad()
loss = model(Y, T)
loss.backward()
optimizer.step()
report_loss += loss.item()
#break
n_samples += len(Y.data)
count += 1
if count % args.report == 0 or count == total:
print('%d/%d, epoch: %d, report_loss: %.3f, time: %.2f' %
(count, total, i + 1, report_loss / n_samples,
time.time() - start_time))
score = eval(model, dev_set, args.batch_size)
model.train()
if score > best_score:
best_score = score
save_model(os.path.join(args.dir, 'best_checkpoint.pt'),
model)
else:
save_model(os.path.join(args.dir, 'checkpoint.pt'), model)
report_loss, start_time, n_samples = 0, time.time(), 0
return model
def train_and_eval_network(config, args):
test_data = TestSet(config["test_data"], Api(API_ENDPOINT))
full_train_data = get_full_train_data(config)
if config["is_existing"] == False:
with DataSet(config["train_data"], Api(API_ENDPOINT)) as train_data:
train(config, train_data, full_train_data, test_data)
models.save_model(config)
# Find the best threshold for the training data
train_data_config = copy.deepcopy(config["test_data"])
train_data_config["patient_filter"] = config["train_data"][
"patient_filter"]
train_data = TestSet(train_data_config, Api(API_ENDPOINT))
config["alert_threshold"], config[
"alert_threshold_low"] = find_best_threshold(config, train_data)
save_config(config)
result = test(config, test_data)
result["default"] = result["alarm_fscore"]
result["threshold"] = config["alert_threshold"]
result["threshold_low"] = config["alert_threshold_low"]
result["tag"] = args["tag"]
if args["results_path"] is not None:
with open(args["results_path"], "a") as file:
json.dump(result, file)
file.write("\n")
return result
def train_composition(dataset, transformation_list):
"""
Train a model on dataset on which a sequence of transformations applied
:param dataset: the original dataset
:param transformation_list: the sequence of transformations
:return:
"""
# Apply a sequence of transformations
(X_train, Y_train), (X_test, Y_test) = load_data(dataset)
X_train = transform(X_train, transformation_list)
nb_examples, img_rows, img_cols, nb_channels = X_train.shape
nb_classes = Y_train.shape[1]
input_shape = (img_rows, img_cols, nb_channels)
# Train a model and save
model_name = 'model-{}-cnn-{}'.format(dataset, 'composition')
require_preprocess = (dataset == DATA.cifar_10)
model = models.create_model(dataset, input_shape, nb_classes)
models.train(model, X_train, Y_train, model_name, require_preprocess)
# save to disk
models.save_model(model, model_name)
# evaluate the new model
loaded_model = models.load_model(model_name)
X_test = transform(X_test, transformation_list)
if require_preprocess:
X_test = normalize(X_test)
scores = loaded_model.evaluate(X_test, Y_test, verbose=2)
print('*** Evaluating the new model: {}'.format(scores))
del loaded_model
def save(self, model_path):
weights_file = os.path.join(model_path, "weights.pkl")
params_file = os.path.join(model_path, "params.pkl")
preprocessor_file = os.path.join(model_path, "preprocessor.pkl")
metadata_file = os.path.join(model_path, "metadata.json")
self.p.save(preprocessor_file)
save_model(self.model, weights_file, params_file)
self.save_metadata(metadata_file)
def train_single(self, model, dataset):
model = model.to(self.opt['device'])
model_optim = optim.Adam(model.parameters(),
lr=self.opt["g_lr"],
betas=(self.opt["beta_1"],
self.opt["beta_2"]))
optim_scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer=model_optim,
milestones=[200, 400, 600, 800],
gamma=self.opt['gamma'])
writer = SummaryWriter(
os.path.join('tensorboard', self.opt['save_name']))
start_time = time.time()
dataloader = torch.utils.data.DataLoader(
dataset=dataset,
shuffle=True,
num_workers=self.opt["num_workers"],
pin_memory=True,
batch_size=self.opt['minibatch'])
L1loss = nn.L1Loss().to(self.opt["device"])
step = 0
for epoch in range(self.opt['epoch_number'], self.opt['epochs']):
self.opt["epoch_number"] = epoch
for batch_num, inout in enumerate(dataloader):
model.zero_grad()
in_coords, out_vals = inout
in_coords = in_coords.to(self.opt['device'])
out_vals = out_vals.to(self.opt['device'])
recovered_data = model(in_coords)
L1 = L1loss(recovered_data, out_vals)
L1.backward()
model_optim.step()
optim_scheduler.step()
psnr = PSNR(recovered_data, out_vals)
if (step % self.opt['save_every'] == 0):
print("Epoch %i batch %i, sf: L1: %0.04f, PSNR (dB): %0.02f" % \
(epoch, batch_num, L1.item(), psnr.item()))
writer.add_scalar('L1', L1.item(), step)
step += 1
if (epoch % self.opt['save_every'] == 0):
save_model(model, self.opt)
print("Saved model")
end_time = time.time()
total_time = start_time - end_time
print("Time to train: " + str(total_time))
save_model(model, self.opt)
print("Saved model")
def run(args):
logging.info('Loading data...')
ds = VarDialDataSet(args.data_directory,
samples_file=args.samples_file,
dialect_labels_file=args.dialect_labels_file,
category_labels_file=args.category_labels_file)
ds.initialize()
samples = ds.samples
labels = ds.dialect_labels
if args.debug:
logging.info(
'Running in debug mode. Dataset is restricted to {} samples.'.
format(args.num_debug_samples))
samples = ds.samples[:args.num_debug_samples]
labels = ds.dialect_labels[:args.num_debug_samples]
samples_train, samples_test, \
labels_train, labels_test = train_test_split(samples, labels)
logging.info('Training vectorizers on text...')
vocab = build_common_vocabulary(samples_train)
ro, md = train_dialect_vectorizers(samples_train, labels_train, vocab)
logging.info('Reshaping input data...')
x = reshape_input_data(ro.transform(samples_train[:1]),
md.transform(samples_train[:1]))
print(x.shape)
logging.info('Building model...')
model = build_dialect_classification_model(x.shape[1:], args.dropout_rate)
print(model.summary())
logging.info('Training the model...')
train_generator = BatchGenerator(samples_train,
labels_train,
ro,
md,
batch_size=args.batch_size)
model.fit_generator(train_generator,
epochs=args.num_epochs,
callbacks=build_model_callbacks())
logging.info('Scoring the model...')
test_generator = BatchGenerator(samples_test,
labels_test,
ro,
md,
batch_size=args.batch_size)
score, acc = model.evaluate_generator(test_generator)
print('Test score: {}'.format(score))
print('Test accuracy: {}.'.format(acc))
logging.info('Saving model and vectorizers...')
save_model(model, ro, md, args.save_model_to)
logging.info("That's all folks!")
def save(mlp, args):
params = [
args.feature_extractor,
str(args.learning_rate),
str(args.epochs),
str(args.alpha),
str(args.num_examples)
]
params.extend([str(l) for l in args.layers])
model_name = 'mlp_' + '_'.join(params)
models.save_model(mlp, model_name)
def on_epoch_end(self, epoch, logs=None):
loss = logs['loss']
validation_loss = logs['val_loss']
self.statistics.loss.append(loss)
self.statistics.validation_loss.append(validation_loss)
save_model(SaveFile(self.training.alphabet, self.training.model),
self.path + '.json',
self.path + '.weights-%s-%s.h5' % (epoch + 1, validation_loss))
save_statistics(self.path + '.statistics.json', self.statistics)
save_config(self.path + '.config.json', self.config)
def train_model(data, model_name):
X, Y = data
transformation_type = TRANSFORMATION.clean
model = models.create_model(DATA.CUR_DATASET_NAME)
print('Training model [{}]...'.format(model_name))
model = models.train(model, X, Y, model_name)
print('Saving model...')
models.save_model(model, model_name)
print('Done.')
return model
def train_cnn(directory, model_path, num_epochs, lr):
"""
Train a Convolutional Neural Network (CNN).
:param directory: directory
:param model_path: model_path
:param num_epochs: number of epochs
:param lr: learning rate
"""
model = CNN()
load_model(model, model_path)
train_on_directory(directory, model, 'CNN', num_epochs, lr)
save_model(model, model_path)
def train_rnn(directory, model_path, num_epochs, lr):
"""
Train a Recurrent Neural Network (RNN).
:param directory: directory
:param model_path: model path
:param num_epochs: number of epochs
:param lr: learning rate
"""
model = RNN(13, 100, 1, 100, 3)
load_model(model, model_path)
train_on_directory(directory, model, 'RNN', num_epochs, lr)
save_model(model, model_path)
def main():
args = init_args()
if args.list_models:
print('\n'.join(models.get_model_names()))
exit()
m = args.model.split(',')
dict_m = models.get_models(m)
x, y = preprocess.load_data(args.train_datesets)
for model_name in dict_m:
model = dict_m[model_name]
print('Training model %s' % model_name)
model.fit(x, y)
models.save_model(model, model_name, args.model_dir)
print('Train finished, save to %s' % args.model_dir)
def fleischner_classification(dataset_path):
print('\nNodule Fleischner Classification\n')
# gpu_devices = tf.config.experimental.list_physical_devices('GPU')
# tf.config.experimental.set_memory_growth(gpu_devices[0], True)
scans, segmentations = dt.load_data(path=dataset_path)
images, labels = dt.process_data_3d(scans,
segmentations,
path=dataset_path,
image_size=64)
(train_images,
train_labels), (test_images,
test_labels) = dt.prepare_data(images,
labels,
should_balance=True)
model, loaded = mdl.load_model('fleischner')
if not loaded:
model = mdl.create_model_fc(input=(64, 64, 64, 1), output=4)
model.summary()
if not loaded:
start_time = time.perf_counter()
history = model.fit(train_images,
train_labels,
batch_size=15,
epochs=60,
validation_split=0.10)
end_time = time.perf_counter()
print('Total time elapsed: {}s'.format(end_time - start_time))
plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label='val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0, 1])
plt.legend(loc='lower right')
plt.show()
mdl.save_model(model, 'fleischner')
score = model.evaluate(test_images, test_labels, verbose=0)
print(model.metrics_names)
print(score)
def on_epoch_end(self, epoch, logs=None):
epoch_str = "%03d" % self.epoch
folder = self.path + epoch_str
if not os.path.exists(folder):
os.mkdir(folder)
model_weights = os.path.join(folder, "weights.pkl")
model_params = os.path.join(folder, "params.pkl")
model_preprocessor = os.path.join(folder, "preprocessor.pkl")
self.preprocessor.save(model_preprocessor)
save_model(self.model, model_weights, model_params)
self.epoch += 1
def main(args):
config = init_for(args["config"])
configure_data_filters(config, [0])
save_config(config)
test_data = TestSet(config["test_data"], Api(API_ENDPOINT))
full_train_data = get_full_train_data(config)
if config["is_existing"] == False:
with DataSet(config["train_data"], Api(API_ENDPOINT)) as train_data:
train(config, train_data, full_train_data, test_data)
models.save_model(config)
test(config, test_data)
def train_model(data, transformation_type=TRANSFORMATION.clean):
X, Y = data
print('Transforming training data set [{}]...'.format(transformation_type))
X = transform(X, transformation_type)
model_name = 'model-{}-cnn-{}'.format(DATA.CUR_DATASET_NAME,
transformation_type)
model = models.create_model(DATA.CUR_DATASET_NAME)
print('Training model [{}]...'.format(model_name))
model = models.train(model, X, Y, model_name)
print('Saving model...')
models.save_model(model, model_name)
print('Done.')
return model
def main():
parser = make_cli_parser()
args = parser.parse_args()
env = gym.make(args.env)
agent = models.load_model(args.env, args.agent)
if not agent:
agent = getattr(agents, args.agent)(env.action_space)
loop(env, agent, args.num, args.epsilon, args.learn)
print()
print("Agent:\n", agent)
if args.save:
models.save_model(args.env, args.agent, agent)
def main():
c = color_codes()
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
try:
net = load_model('/home/mariano/Desktop/test.tf')
except IOError:
x = Input([784])
x_image = Reshape([28, 28, 1])(x)
x_conv1 = Conv(filters=32,
kernel_size=(5, 5),
activation='relu',
padding='same')(x_image)
h_pool1 = MaxPool((2, 2), padding='same')(x_conv1)
h_conv2 = Conv(filters=64,
kernel_size=(5, 5),
activation='relu',
padding='same')(h_pool1)
h_pool2 = MaxPool((2, 2), padding='same')(h_conv2)
h_fc1 = Dense(1024, activation='relu')(h_pool2)
h_drop = Dropout(0.5)(h_fc1)
y_conv = Dense(10)(h_drop)
net = Model(x,
y_conv,
optimizer='adam',
loss='categorical_cross_entropy',
metrics='accuracy')
print(c['c'] + '[' + strftime("%H:%M:%S") + '] ' + c['g'] + c['b'] +
'Original (MNIST)' + c['nc'] + c['g'] + ' net ' + c['nc'] + c['b'] +
'(%d parameters)' % net.count_trainable_parameters() + c['nc'])
net.fit(mnist.train.images,
mnist.train.labels,
val_data=mnist.test.images,
val_labels=mnist.test.labels,
patience=10,
epochs=200,
batch_size=1024)
save_model(net, '/home/mariano/Desktop/test.tf')
def main():
args = parse_args()
X = read_smiles(args.compounds)
Y = load_labels(args.targets)
if not isinstance(Y, np.ndarray):
Y = Y.A
model = build_model(args)
model_dir = os.path.join(args.path, )
os.makedirs(model_dir, exist_ok=True)
model_filename = os.path.join(model_dir,
get_model_filename(args))
if not os.path.exists(model_filename):
print ("fitting PPB2 model")
model.fit(X, Y)
save_model(model, model_filename)
def run_tests(model_name=None):
global epochs
global env
global ac
env = gym.make('StrategyEnv-v0')
test_tf = '20200201-20200207'
# test_tf = '20200401-'
env.set_timeframe(test_tf)
env.full_reset()
def make_env():
env = gym.make('StrategyEnv-v0')
env.set_timeframe('20191110-20200131')
env.randomize_timeframe(True)
env.set_ac(True)
env.full_reset()
return env
#ask joey to run normal BT to compare
run_test_with_strat(env)
env.run_normal_bt()
if model_name:
torch.manual_seed(10000)
np.random.seed(10000)
ac = models.load_model(model_name, env.observation_space,
env.action_space)
else:
# ac = ppo(make_env, epochs=epochs, target_kl=0.001, steps_per_epoch=7200, max_ep_len=100000)
ac = ppo(make_env,
epochs=epochs,
steps_per_epoch=7200,
max_ep_len=100000)
model_name = models.save_model(ac)
run_model_test(env, ac, model_name)
def train_model(model, optimizer, lr_scheduler, dataloaders, start_epoch,
end_epoch, best_acc, best_epoch, best_model, device, logger,
phase, epochs_per_save, save_path, augment_epoch,
unaugment_epoch):
start_time_total = time.time()
saved_epoch = -1
for epoch in range(start_epoch + 1, end_epoch + 1):
print(f'Epoch {epoch}/{end_epoch}:', end='\t')
# Each epoch has a training and validation phase
train(model, optimizer, lr_scheduler, dataloaders["train"], device,
logger, epoch)
epoch_acc = test(model, dataloaders["test"], device, logger, epoch)
if epoch_acc > best_acc:
print("Best so far")
best_acc = epoch_acc
best_epoch = epoch
best_model = copy.deepcopy(model.state_dict())
else:
print(f'Best epoch: {best_epoch} : {best_acc:4f}')
## fine tuning with unaugmented dataset ##
if phase > 0 and epoch % augment_epoch == 0:
start_time = time.time()
print("Fine Tuning with an unaugmented dataset")
dataloader = dataloaders['train']
dataloader.set_resolution("unaugmented")
model.train()
epoch_loss, epoch_acc, total_data_size = 0.0, 0, 0
iter_end, iteration = len(dataloader) * unaugment_epoch, 0
while iteration < iter_end:
for inputs, labels in dataloader:
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = cross_entropy(outputs, labels)
loss.backward()
optimizer.step()
_, preds = torch.max(outputs, 1)
_, labels = torch.max(labels, 1)
iteration += 1
epoch_loss += loss.item() * inputs.size(0)
epoch_acc += torch.sum(preds == labels).item()
total_data_size += inputs.size(0)
if iteration >= iter_end:
break
epoch_loss /= total_data_size
epoch_acc /= total_data_size
logger.log("train", epoch + 0.5, epoch_loss, epoch_acc)
end_time = time.time()
print(
f'\t\tTrain Loss: {epoch_loss:.4f}\tAcc: {epoch_acc:.4f}\tTime: {end_time - start_time:.2f}sec'
)
dataloader.set_resolution("augmented")
# test fine-tuned model
epoch_acc = test(model, dataloaders['test'], device, logger,
epoch + 0.5)
if epoch_acc > best_acc:
print("Best so far")
best_acc = epoch_acc
best_epoch = epoch
best_model = copy.deepcopy(model.state_dict())
else:
print(f'Best epoch: {best_epoch} : {best_acc:4f}')
if epoch % epochs_per_save == 0:
save_model(model, optimizer, lr_scheduler, epoch, phase, best_acc,
best_epoch, best_model, save_path)
saved_epoch = epoch
if epoch > saved_epoch:
save_model(model, optimizer, lr_scheduler, epoch, phase, best_acc,
best_epoch, best_model, save_path)
sec = time.time() - start_time_total
print(
f'====Total time: {int(sec)//3600}h {(int(sec)%3600)//60}m {sec%60:.2f}s ({sec:.2f} s)===='
)
# load best model weights
model.load_state_dict(best_model)
return model
display_predictions(convert_to_color(train_gt), viz, caption="Train ground truth")
display_predictions(convert_to_color(test_gt), viz, caption="Test ground truth")
if MODEL == 'SVM_grid':
print("Running a grid search SVM")
# Grid search SVM (linear and RBF)
X_train, y_train = build_dataset(img, train_gt,
ignored_labels=IGNORED_LABELS)
class_weight = 'balanced' if CLASS_BALANCING else None
clf = sklearn.svm.SVC(class_weight=class_weight)
clf = sklearn.model_selection.GridSearchCV(clf, SVM_GRID_PARAMS, verbose=5, n_jobs=4)
clf.fit(X_train, y_train)
print("SVM best parameters : {}".format(clf.best_params_))
prediction = clf.predict(img.reshape(-1, N_BANDS))
save_model(clf, MODEL, DATASET)
prediction = prediction.reshape(img.shape[:2])
elif MODEL == 'SVM':
X_train, y_train = build_dataset(img, train_gt,
ignored_labels=IGNORED_LABELS)
class_weight = 'balanced' if CLASS_BALANCING else None
clf = sklearn.svm.SVC(class_weight=class_weight)
clf.fit(X_train, y_train)
save_model(clf, MODEL, DATASET)
prediction = clf.predict(img.reshape(-1, N_BANDS))
prediction = prediction.reshape(img.shape[:2])
elif MODEL == 'SGD':
X_train, y_train = build_dataset(img, train_gt,
ignored_labels=IGNORED_LABELS)
X_train, y_train = sklearn.utils.shuffle(X_train, y_train)
scaler = sklearn.preprocessing.StandardScaler()
"val": [it],
"predict": [it]})
if opt.predict:
trainer.predict_masks(loaders["predict"], "/media/nick/DATA/ame_predicts/", it)
del loaders
continue
if lr_schedule:
for i in range(start_epoch - 1):
lr_schedule.step()
print([group_param["lr"] for group_param in optimizer.param_groups])
for epoch in range(start_epoch + 1, opt.num_epochs + 1):
log_dict_val, log_dict_test = None, None
log_dict_train = trainer.train(epoch, loaders["train"])
save_model(os.path.join(opt.save_dir, 'model_last.pth'),
epoch, model, -1, optimizer)
if "val" in loaders and opt.val_intervals > 0 and not (epoch % opt.val_intervals):
with torch.no_grad():
log_dict_val = trainer.val(epoch, loaders["val"])
need_save, timespamp = history.step(epoch, log_dict_train, log_dict_val, log_dict_test)
if need_save:
save_model(os.path.join(opt.save_dir, str(timespamp) + '.pth'),
epoch, model, log_dict_train["loss"], optimizer)
if lr_schedule:
lr_schedule.step()
print([group_param["lr"] for group_param in optimizer.param_groups])
os.rename(os.path.join(opt.save_dir, 'model_last.pth'), os.path.join(opt.save_dir, f'model_last{it}.pth'))
os.rename(os.path.join(opt.save_dir, 'model_best.pth'), os.path.join(opt.save_dir, f'model_best{it}.pth'))
del loaders
CLASS_NAMES = ["dog", "cat"]
NUM_TRAIN_EPOCHS = 10
STOP_IF_NOT_IMPROVE = 0
LOAD_MODEL = True
LOAD_MODEL_PATH = 'models/model_1574546050.9673572.pkl'
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if __name__ == '__main__':
torch.multiprocessing.freeze_support()
if REORGANIZED_DATA:
d = DataOrganization()
d.organized_data(IMG_SIZE, VAL_DIR_NAME, TRAIN_DIR_NAME, CLASS_NAMES,
DATA_ROOT_DIR)
data_loaders = create_dataloders(DATA_ROOT_DIR)
net, epoch = get_model(LOAD_MODEL, LOAD_MODEL_PATH, IMG_SIZE,
len(CLASS_NAMES))
net, best_acc, best_loss, accs_train, losses_train, accs_val, losses_val = \
train_and_eval(net, data_loaders, NUM_TRAIN_EPOCHS, STOP_IF_NOT_IMPROVE)
save_model(net, best_acc, epoch + len(accs_val))
# plot_train_results(best_acc, best_loss, accs_train, losses_train, accs_val, losses_val)
test(net, os.path.join(VAL_DIR_NAME, CLASS_NAMES[0]),
data_loaders['train'].dataset.class_to_idx, 10)
test(net, os.path.join(VAL_DIR_NAME, CLASS_NAMES[1]),
data_loaders['train'].dataset.class_to_idx, 10)
def save(self, weights_file, params_file, preprocessor_file):
self.p.save(preprocessor_file)
save_model(self.model, weights_file, params_file)
def save_model(self, path='./checkpoints', postfix=None):
save_model(self.model, path, postfix)
def trainBoostedClassifier(self, classifier, level=0):
milestones = self.config['milestones']
lr = self.config['lr']
if self.resume_epoch != -1:
self.classifiers = self.classifiers[:-1]
self.weights = self.weights[:-1]
start = self.resume_epoch
tmp = -1
for m in range(len(milestones)):
if milestones[m] <= self.resume_epoch:
lr = lr * self.config['gamma']
tmp = m
else:
break
if tmp != -1:
milestones = milestones[tmp:]
milestones = list(np.array(milestones) - self.resume_epoch)
else:
start = 0
id_classifier = len(self.classifiers)
print(level * " " + "Training Boosted Classifier n°" +
str(id_classifier) + "...")
optimizer = optim.SGD(classifier.parameters(),
lr=lr,
momentum=self.config['momentum'],
weight_decay=self.config['weight_decay'])
scheduler = lr_scheduler.MultiStepLR(optimizer,
milestones=milestones,
gamma=self.config['gamma'],
last_epoch=-1)
#Adversarial training for the first classifier
if id_classifier == 0:
attack = LinfPGDAttack(classifier,
eps=self.config['eps'] / 255,
eps_iter=self.config['eps_iter'] / 255,
nb_iter=self.config['nb_iter'],
rand_init=self.config['rand_init'],
clip_min=self.config['clip_min'],
clip_max=self.config['clip_max'])
for epoch in range(start, self.config['epochs']):
classifier.train()
models.adversarialTrain(classifier,
self.device,
self.train_loader,
optimizer,
epoch,
attack,
level=level + 1)
scheduler.step()
classifier.eval()
accuracy_under_attack = models.test_under_attack(
classifier,
self.device,
self.test_loader,
attack,
level=level + 1)
accuracy = models.test(classifier,
self.device,
self.test_loader,
level=level + 1)
models.save_model(self.save_dir,
id_classifier,
self.device,
classifier,
accuracy,
accuracy_under_attack,
epoch,
level=level + 1)
models.updateAndSaveBestAccuracies(self.save_dir,
id_classifier,
self.device,
classifier,
accuracy,
accuracy_under_attack,
level=level + 1)
else: #Natural training on the adversarial data set created against the mixture
adversarial_train_loader, adversarial_test_loader = self.adversarialTrainLoader(
level=level + 1)
for epoch in range(start, self.config['epochs']):
classifier.train()
models.train(classifier,
self.device,
adversarial_train_loader,
optimizer,
epoch,
level=level + 1)
scheduler.step()
classifier.eval()
accuracy_under_attack = models.test(classifier,
self.device,
adversarial_test_loader,
level=level + 1)
accuracy = models.test(classifier,
self.device,
self.test_loader,
level=level + 1)
models.save_model(self.save_dir,
id_classifier,
self.device,
classifier,
accuracy,
accuracy_under_attack,
epoch,
level=level + 1)
models.updateAndSaveBestAccuracies(self.save_dir,
id_classifier,
self.device,
classifier,
accuracy,
accuracy_under_attack,
level=level + 1)
classifier, acc, top_acc_under_attack = models.load_model(
self.save_dir,
id_classifier,
-1,
self.device,
self.config['number_of_class'],
top_acc_under_attack=True)
self.classifiers.append(classifier)