def train(cfg):
print('Preparing to train {} on {} data...'.format(cfg['main']['model_name'],
cfg['main']['dataset_name']))
np.random.seed(1337) # for reproducibility
print('Tensorflow backend detected; Applying memory usage constraints')
ss = K.tf.Session(config=K.tf.ConfigProto(gpu_options=K.tf.GPUOptions(allow_growth=True)))
K.set_session(ss)
ss.run(K.tf.global_variables_initializer())
K.set_learning_phase(1)
Prostate_dataset = select_dataset(cfg)
if not cfg["validation"]["do_validation"]:
image_gen = Prostate_dataset.create_generator()
val_data = None
else:
image_gen, val_data = Prostate_dataset.create_generator_with_validation()
model = select_model(cfg, Prostate_dataset)
# model.summary()
myAdam = keras.optimizers.Adam(lr=1e-4)
model.compile(optimizer=Adam(lr=1e-4),
loss={'Radboud_Branch': 'categorical_crossentropy', 'Karolinska_Branch': 'categorical_crossentropy'}, loss_weights=[0.5, 0.5], metrics=['accuracy'])
print('Created image generator from dataset {}!'.format(cfg['main']['dataset_name']))
tf_saver = tf.train.Saver(max_to_keep=2)
checkpoint_callback = TFCheckPointCallback(tf_saver, ss, cfg)
modelsaver_callback = SaveKerasModel(cfg)
learningRate_callback = SelectLRCallBack(cfg)
tbCallBack = TensoBoardCallback(cfg)
model.fit_generator(image_gen,
steps_per_epoch=Prostate_dataset.sample_number / cfg['training']['batch_size'],
epochs=cfg['training']['epochs'],
callbacks=[checkpoint_callback, learningRate_callback, modelsaver_callback, tbCallBack],
validation_data=val_data)
args.o2n = o2n
args.n2o = n2o
for key in ['train_labels_path', 'val_labels_path']:
if args[key] == 'None':
args[key] = None
settings.set_settings(args)
#args.output_path = tempfile.TemporaryDirectory().name
os.makedirs(args.output_path)
train_loader, val_loader, labelled_train_loader = dataset.get_data_loaders(
args)
model = models.select_model(args)
def SIDX(template_id):
return (3 + (template_id - 1) * 7)
def EIDX(template_id):
return (3 + (template_id - 1) * 7 + 7)
# for i in val_loader.dataset.raw_data[:,SIDX(6)]:
# print (i)
# print (val_loader.dataset.raw_data[:,SIDX(2)])
#my_score, max_score, similarity, rank, conditional_rank, mean, std
ss = K.tf.Session(config=K.tf.ConfigProto(gpu_options=K.tf.GPUOptions(allow_growth=True)))
K.set_session(ss)
# debugging parameters
interim_testing = False
# parameters
dataset_name = 'mscoco'
load_mscoco = False
dw = 512
dh = 512
model_name = 'enet_unpooling'
pw = os.path.join('models', dataset_name, 'enet_unpooling', 'weights', '{}_best.h5'.format(model_name))
nc = datasets.load(dataset_name).num_classes()
autoencoder = models.select_model(model_name=model_name)
segmenter, model_name = autoencoder.build(nc=nc, w=dw, h=dh)
segmenter.load_weights(pw)
if load_mscoco:
data = load_mscoco_data(segmenter=segmenter)
else:
txt_file = sys.argv[1]
image_dir = os.path.dirname(txt_file)
with open(txt_file) as fin:
image_filenames = [os.path.join(image_dir, line.rstrip('\n')) for line in fin]
data = load_arbitrary_data(segmenter=segmenter, image_filenames=image_filenames)
data_gen = data['data_gen']
if interim_testing:
for idx, item in enumerate(data_gen):
def main():
# build parser and check arguments
args = _build_parser()
_check_args(args)
# Setup Estimator
'''Estimator name:
xgb: XGBoost Classifier
log: Logistic Regression
knn: KNeighbors Classifier
rfo: RandomForest Classifier
ada: AdaBoost Classifier
ext: ExtraTrees Classifier
svc: Support Vector Classifier
keras: Keras Neural Networks
'''
if not args.estimator == 'all':
estimators = [args.estimator]
elif args.estimator == 'all':
estimators = ['xgb', 'lgb', 'log', 'rfo', 'ext', 'ada', 'knn', 'svc']
# Training neural nets with keras
if args.train_nn:
estimator_name = 'keras'
print('Training %s...' % estimator_name)
params = {
'n_features': n_features,
'n_classes': n_classes,
'dropout': args.dropout,
'hidden_unit': args.hidden_unit,
'n_layers': args.layers,
'optimizer': args.optimizer,
'init': args.init,
'batch_size': args.batch_size,
'epochs': args.epochs,
}
estimator = keras_model(**params)
train_kwargs = {
'X_train': X_train,
'y_train': y_train,
'X_val': X_val,
'y_val': y_val,
'score_name': args.score,
'num': args.num
}
_ = estimator.train(**train_kwargs)
print('params: \n', params)
# Training random search CV with scikit-learn models
if args.train_random:
for estimator_name in estimators:
print('Training %s...' % estimator_name)
if not estimator_name == 'keras':
seed = args.seed if args.seed != None else np.random.randint(
100)
estimator, params = select_model(estimator_name, n_features,
n_classes, seed)
# kwargs dict for train and predict
train_kwargs = {
'estimator': estimator,
'params': params,
'X_train': X_train,
'y_train': y_train,
'X_val': X_val,
'y_val': y_val,
'n_iter': args.n_iter,
'score_name': args.score,
}
# Train model and Predict results
best_params, best_score, val_score = random_model(
**train_kwargs)
timestamp = get_timestamp()
# Write params to file
write_params(estimator_name, best_params, best_score,
val_score, timestamp, args.num)
elif estimator_name == 'keras':
space_params = {
'n_features':
n_features,
'n_classes':
n_classes,
'dropout':
hp.uniform('dropout', .20, .80),
'hidden_unit':
hp.quniform('hidden_unit', 10, 50, q=1),
'n_layers':
hp.choice('n_layers', [1, 2, 3, 4]),
'optimizer':
hp.choice('optimizer', ['adam', 'adadelta', 'sgd']),
'init':
hp.choice('init', ['glorot_uniform', 'normal', 'uniform']),
'batch_size':
hp.choice('batch_size', [16, 32, 64, 128]),
'epochs':
hp.quniform('epochs', 100, 1000, q=1),
'score_name':
args.score,
'num':
args.num,
}
trials = Trials()
best_params = fmin(random_nn,
space_params,
algo=tpe.suggest,
max_evals=args.n_iter,
trials=trials)
print('best_params \n', best_params)
# Evaluate with ensemble method and predict result
if args.predict:
eva_kwargs = {
'estimators': estimators,
'threshold': args.threshold,
'X_train': X_train,
'y_train': y_train,
'X_val': X_val,
'y_val': y_val,
'X_test': X_test,
'score_name': args.score,
'n_classes': n_classes,
}
# Predict with ensemble voting and write result
prediction = ensemble(**eva_kwargs)
if args.ensemble == 'vote':
result = prediction.vote()
elif args.ensemble == 'stack':
result = prediction.stack(args.num_imp)
timestamp = get_timestamp()
write_result(result, label_list, timestamp)
def main(args):
# Store name of experiment
exp_name = args.exp_name
exp_name = '{}_r{}_p{}_n{}_i{}_k{}'.format(exp_name, args.rho,
args.pos_reward,
args.neg_reward,
args.class_imbalance,
args.kldiv_lambda)
# Create an directory for output path
args.output_path = os.path.join(args.output_path, args.exp_name)
os.makedirs(args.output_path, exist_ok=True)
utils.LOG_FILE = os.path.join(args.output_path, 'log.txt')
LEARNING_PROFILE_FILE = os.path.join(args.output_path,
'learning_curve.txt')
lpf = open(LEARNING_PROFILE_FILE, 'a')
args.lpf = lpf
# Set logging
logging.basicConfig(filename=utils.LOG_FILE,
filemode='a',
format='%(levelname)s :: %(asctime)s - %(message)s',
level=args.log_level,
datefmt='%d/%m/%Y %I:%M:%S %p')
console = logging.StreamHandler()
console.setLevel(args.log_level)
formatter = logging.Formatter('%(levelname)s :: %(asctime)s - %(message)s',
datefmt='%d/%m/%Y %I:%M:%S %p')
console.setFormatter(formatter)
logging.getLogger().addHandler(console)
logging.info(
'Beginning code for experiment {} and storing stuff in {}'.format(
exp_name, args.output_path))
logging.info('Loaded arguments as \n{}'.format(str(pprint.pformat(args))))
# Begin of main code
train_loader, val_loader, labelled_train_loader = dataset.get_data_loaders(
args)
model = models.select_model(args)
my_eval_fn = compute.get_evaluation_function(args)
if args.optim == 'sgd':
optimizer = optim.SGD(filter(lambda p: p.requires_grad,
model.parameters()),
momentum=args.momentum,
lr=args.lr,
weight_decay=args.decay)
else:
optimizer = optim.Adam(filter(lambda p: p.requires_grad,
model.parameters()),
lr=args.lr,
weight_decay=args.decay)
checkpoint_file = os.path.join(args.output_path,
'{}_checkpoint.pth'.format(exp_name))
best_checkpoint_file = os.path.join(
args.output_path, '{}_best_checkpoint.pth'.format(exp_name))
logging.info('Saving checkpoints at {} and best checkpoint at : {}'.format(
checkpoint_file, best_checkpoint_file))
start_epoch = 0
best_score = -9999999
# Load checkpoint if present in input arguments
if args.checkpoint != '':
logging.info('Starting from checkpoint: {}'.format(args.checkpoint))
cp = torch.load(args.checkpoint)
start_epoch = cp['epoch'] + 1
model.load_state_dict(cp['model'])
# optimizer.load_state_dict(cp['optimizer']) TODO: - Why not do this?
best_score = cp['best_score']
for param_group in optimizer.param_groups:
param_group['lr'] = args.lr
param_group['weight_decay'] = args.decay
num_epochs = args.num_epochs
logging.info('Beginning train/validate cycle')
time1 = time.time()
if val_loader is not None:
record, metric_idx, headers = compute.compute(start_epoch - 1,
model,
val_loader,
optimizer,
'eval',
eval_fn=my_eval_fn,
args=args)
if (args.log_eval is not None):
handler = open(args.log_eval, "a")
print(','.join([
str(round(x, 6)) if isinstance(x, float) else str(x)
for x in record
]),
file=handler)
handler.close()
print("Time taken:", time.time() - time1)
if (args.only_eval):
logging.info('Ran only eval mode, now exiting')
exit(0)
# Start TRAINING
for epoch in range(start_epoch, num_epochs):
logging.info('Beginning epoch {}'.format(epoch))
if labelled_train_loader is not None:
record, metric_idx, _ = compute.compute(epoch,
model,
labelled_train_loader,
optimizer,
'train_sup',
eval_fn=my_eval_fn,
args=args)
if train_loader is not None:
record, metric_idx, _ = compute.compute(
epoch,
model,
train_loader,
optimizer,
'train_un',
eval_fn=my_eval_fn,
args=args,
labelled_train_loader=labelled_train_loader)
if val_loader is not None:
record, metric_idx, _ = compute.compute(epoch,
model,
val_loader,
None,
'eval',
eval_fn=my_eval_fn,
args=args)
is_best = False
logging.info('Best score: {}, This score: {}'.format(
best_score, record[metric_idx]))
if record[metric_idx] > best_score:
best_score = record[metric_idx]
is_best = True
utils.save_checkpoint(
{
'epoch': epoch,
'best_score': best_score,
'model': model.state_dict(),
'optimizer': optimizer.state_dict(),
'is_best': is_best
}, epoch, is_best, checkpoint_file, best_checkpoint_file)
args.lpf.close()
def __init__(self,config,train_loader,val_loader,model_name):
self.config = config
self.train_loader = train_loader
self.val_loader = val_loader
self.model_name = model_name
self.num_train = len(self.train_loader.dataset)
self.num_valid = len(self.val_loader.dataset)
self.saved_model_dir = self.config.saved_model_dir
self.output_model_dir = self.config.output_model_dir
self.best_model = config.best_model
self.use_gpu = self.config.use_gpu
if(self.config.resume == True):
print("LOADING SAVED MODEL")
self.net = select_model(self.model_name,self.config.file_type)
self.loadCheckpoint()
else:
print("INTIALIZING NEW MODEL")
self.net = select_model(self.model_name,self.config.file_type)
self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
self.net = self.net.to(self.device)
self.total_epochs = config.epochs
if(self.model_name =="lstm_model" or self.model_name =="cnn_lstm_model"):
print("NLL LOSS")
self.criterion = nn.NLLLoss()
self.optimizer = optim.Adam(self.net.parameters(), lr=0.0001)
loss_name = "nll_loss"
lr = 0.0001
else:
self.criterion = nn.CrossEntropyLoss()
self.optimizer = optim.SGD(self.net.parameters(),lr = self.config.lr,
momentum = self.config.momentum,
weight_decay = self.config.weight_decay)
loss_name = "crossentropy_loss"
lr = self.config.lr
self.num_params = sum([p.data.nelement() for p in self.net.parameters()])
self.batch_size = self.config.batch_size
self.train_paitence = config.train_paitence
self.num_classes = 10
if(self.config.debug == False):
self.experiment = wandb.init(project="audio_classification")
hyper_params = {
"model_name": self.model_name,
"file_type": self.config.file_type,
"dataset": self.config.dataset,
"batch_size": self.config.batch_size,
"num_epochs": self.total_epochs,
"loss_function": loss_name,
"learning_rate": lr,
"momentum": self.config.momentum,
"weight_decay": self.config.weight_decay
}
self.experiment.config.update(hyper_params)
wandb.watch(self.net)
# summary(self.net, input_size=(1, 128, 216))
print('[*] Number of model parameters: {:,}'.format(self.num_params))