def __init__(self, modeluse):
from utils.utils import load_model
from modeling.TP_Net import Res160, Res320
from modeling.Hourglass import HourglassNet
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if device == 'cpu':
raise EOFError('cpu version for training is not implemented.')
print('Using device: ', device)
self.head = {'center': 1, 'dis': 4, 'line': 1}
if modeluse == 'tp320':
self.model = load_model(Res320(self.head),
'./pretraineds/Res320.pth')
self.in_res = (320, 320)
elif modeluse == 'tplite':
self.model = load_model(Res160(self.head),
'./pretraineds/Res160.pth')
self.in_res = (320, 320)
elif modeluse == 'tp512':
self.model = load_model(Res320(self.head),
'./pretraineds/Res512.pth')
self.in_res = (512, 512)
elif modeluse == 'hg':
self.model = load_model(HourglassNet(self.head),
'./pretraineds/HG128.pth')
self.in_res = (512, 512)
else:
raise EOFError(
'Please appoint the correct model (option: tp320, tplite, tp512, hg). '
)
self.model = self.model.cuda().eval()
def tester(self):
(X_test, y_test, mask_test, mem_test, unk_test) = self.testdata
best_models = []
for i in range(1,5):
model = self.Model(self.args).to(self.args.device)
load_model(model, self.args, index=i)
model.eval()
best_models.append(model)
test_loss, confusion_test, confusion_mem_test, (alphas, targets, seq_lengths) = self.run_test(best_models, X_test, y_test, mask_test, mem_test, unk_test)
print("ENSAMBLE TEST RESULTS")
print(confusion_test)
print(confusion_mem_test)
print("test accuracy:\t\t{:.2f} %".format(confusion_test.accuracy() * 100))
print("test Gorodkin:\t\t{:.4f}".format(gorodkin(confusion_test.ret_mat())))
print("test mem accuracy:\t{:.2f} %".format(confusion_mem_test.accuracy() * 100))
print("test mem MCC:\t\t{:.4f}".format(confusion_mem_test.MCC()))
self.results.set_final(
alph = alphas.cpu().detach().numpy(),
seq_len = seq_lengths.cpu().detach().numpy(),
targets = targets,
cf = confusion_test.ret_mat(),
cf_mem = confusion_mem_test.ret_mat(),
acc = confusion_test.accuracy(),
acc_mem = confusion_mem_test.accuracy())
save_results(self.results,self.args)
def main(config):
with tf.Graph().as_default():
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=config.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
with sess.as_default():
detector = MTCNN(config, sess)
# Load the model
load_model(config.lfw.valid_model_path)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
phase_train_placeholder = tf.get_default_graph(
).get_tensor_by_name("phase_train:0")
# Run forward pass to calculate embeddings
with open(config.parse_path) as f:
names = f.readlines()
for name in names:
keyword, name, saved_dir = get_keyword(name, config.output_dir)
fanye_url = url_init_first + urllib.quote(keyword, safe='/')
feats = []
page = 0
while len(os.listdir(saved_dir)) < 20 and fanye_url != '':
onepage_urls, fanye_url = get_onepage_urls(fanye_url)
for index, url in enumerate(onepage_urls):
img_path = download(url,
page * len(onepage_urls) + index,
saved_dir)
try:
img = misc.imread(img_path)
nrof_faces, face = detect(img, detector, config)
face = np.expand_dims(face, 0)
if nrof_faces:
feed_dict = {
images_placeholder: face,
phase_train_placeholder: False
}
feats.append(
sess.run(embeddings, feed_dict=feed_dict))
if index != 0:
dist = np.sqrt(
np.sum(
np.square(
np.subtract(
feats[0], feats[-1]))))
print('the %d-th %s image dist: %f' %
(index, name, dist))
if dist > config.max_dist:
os.remove(img_path)
except Exception as e:
print(e)
if img_path:
os.remove(img_path)
continue
page += 1
def tester(self):
model = self.Model(self.args).to(self.args.device)
load_model(model, self.args)
test_loss, test_accuracy = self.run_test(model=model)
print('| Test | loss {:.4f} | acc {:.2f}%'
' |'.format(test_loss, test_accuracy * 100))
print('-' * 79)
def main(config):
with tf.Graph().as_default():
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=config.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
with sess.as_default():
pairs = lfw.read_pairs(
os.path.expanduser(config.lfw.lfw_pairs_path)
) # Read the file containing the pairs used for testing
paths, actual_issame = lfw.get_paths(
config.input_dir, pairs,
'png') # Get the paths for the corresponding images
load_model(config.lfw.valid_model_path) # Load the model
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
phase_train_placeholder = tf.get_default_graph(
).get_tensor_by_name("phase_train:0")
image_size = config.lfw.image_size
embedding_size = embeddings.get_shape()[1]
print('Runnning forward pass on LFW images')
batch_size = config.batch_size
nrof_images = len(paths)
nrof_batches = int(math.ceil(1.0 * nrof_images / batch_size))
emb_array = np.zeros((nrof_images, embedding_size))
for i in range(nrof_batches):
start_index = i * batch_size
end_index = min((i + 1) * batch_size, nrof_images)
paths_batch = paths[start_index:end_index]
images = load_data(paths_batch, image_size, False, False)
feed_dict = {
images_placeholder: images,
phase_train_placeholder: False
}
emb_array[start_index:end_index, :] = sess.run(
embeddings, feed_dict=feed_dict)
tpr, fpr, accuracy, val, val_std, far = lfw.evaluate(
emb_array, actual_issame, nrof_folds=config.lfw.lfw_nrof_folds)
print('Accuracy: %1.3f+-%1.3f' %
(np.mean(accuracy), np.std(accuracy)))
print('Validation rate: %2.5f+-%2.5f @ FAR=%2.5f' %
(val, val_std, far))
auc = metrics.auc(fpr, tpr)
print('Area Under Curve (AUC): %1.3f' % auc)
eer = brentq(lambda x: 1. - x - interpolate.interp1d(fpr, tpr)(x),
0., 1.)
print('Equal Error Rate (EER): %1.3f' % eer)
def __init__(self, model_path, confidence_threshold=0.7):
self.net = get_hourglass(cfg.arch,
num_classes=cfg.num_classes,
is_training=False)
load_model(self.net, model_path)
self.net = self.net.to(cfg.device)
self.net.eval()
self.nms_threshold = 0.4
self.confidence_threshold = confidence_threshold
def play(self, load_model=True, test_ep=None, num_step=100000, num_episodes=200, display=True):
model_q = Model()
model_target_q = Model(is_target_q=True)
end_points_q = model_q.model_def(self.inputs, self.env, name='main_q')
_ = model_target_q.model_def(
self.target_inputs, self.env, name='target_q')
init = tf.global_variables_initializer()
self.saver = tf.train.Saver(max_to_keep=None)
if load_model:
utils.load_model(self.saver, self.sess, self.model_dir)
else:
self.sess.run(init)
if test_ep is None:
test_ep = self.cfg.ep_test
if not display:
gym_dir = '/tmp/%s-%s' % (self.cfg.env_name, utils.get_time())
self.env.env.monitor.start(gym_dir)
best_reward, best_episode = 0, 0
for episode in xrange(num_episodes):
screen, reward, action, terminal = self.env.new_random_game()
current_reward = 0
for _ in xrange(self.cfg.history_length):
self.history.add(screen)
for t in tqdm(xrange(num_step), ncols=70):
# 1. predict
action = self.predict(
end_points_q['pred_action'], self.history.get(), ep=test_ep)
# 2. act
screen, reward, terminal = self.env.act(
action, is_training=False)
# 3. observe
self.history.add(screen)
current_reward += reward
if terminal:
break
if current_reward > best_reward:
best_reward = current_reward
best_episode = episode
print " [%d] Best reward : %d" % (best_episode, best_reward)
if not display:
self.env.env.monitor.close()
def load_model(self, model_file):
data = load_model(model_file)
self.word2id = data['word2id']
self.label2id = data['label2id']
self.initial_proba = data['initial_proba']
self.observation_proba = data['observation_proba']
self.transition_proba = data['transition_proba']
def main(arguments):
model_parameters = {
'scale': arguments['scale'],
'learning_rate': 1e-5,
'D': arguments['D'],
'C': arguments['C'],
'G': arguments['G'],
'kernel_size': 3,
'c_dim': 3,
'G0': arguments['G0'],
}
model = load_model(model_parameters,
arguments['vgg'],
verbose=arguments['verbose'])
if arguments['summary'] is True:
model.rdn.summary()
if arguments['train'] is True:
from trainer.train import Trainer
trainer = Trainer(train_arguments=arguments)
trainer.train_model(model)
if arguments['test'] is True:
from predict.predict import Predictor
predictor = Predictor(test_arguments=arguments)
predictor.get_predictions(model)
def train_epoch(model, optimizer, args, hp, step, start_epoch=0, rank=0):
dataloader = get_dataloader(step, args, hp)
for epoch in range(start_epoch, hp.max_epoch):
start_time = time.time()
if step // hp.accum_grad > hp.warmup_step:
dataloader = get_dataloader(step, args, hp)
step = train_loop(model, optimizer, step, epoch, args, hp, rank, dataloader)
print("EPOCH {} end".format(epoch+1))
print('elapsed time = {}'.format(time.time() - start_time))
if hp.dev_script is not None:
if (epoch >= 20 and epoch % 5 == 0) or epoch == 1 and rank == 0:
print('eval')
map_location = {'cuda:%d' % 0: 'cuda:%d' % rank}
loaded_dict = load_model("{}".format(os.path.join(hp.save_dir, 'network.epoch{}'.format(epoch+1))), map_location=map_location)
model_dev = Transformer(hp)
model_dev.load_state_dict(loaded_dict)
model_dev.eval()
model_dev.to(rank)
wer_all = recognize(hp, model_dev, model_lm=None, lm_weight=0.0, calc_wer=True, rank=rank)
print(f'{args.hp_file} epoch {epoch} wer is {wer_all}')
del model_dev
def _test(parameters):
model_folder = setup_log(parameters, parameters['save_pred'] + '_test')
print('\nLoading mappings ...')
train_loader = load_mappings(parameters['remodelfile'])
flag=True
print('\nLoading testing data ...')
processed_dataset=parameters['remodelfile']
if flag and os.path.exists(os.path.join(processed_dataset, 'test_test_data.pkl')):
with open(os.path.join(processed_dataset, 'test_test_data.pkl'), 'rb') as f:
test_data = pkl.load(f)
with open(os.path.join(processed_dataset, 'test_prune_recall.pkl'), 'rb') as f:
prune_recall = pkl.load(f)
else:
test_loader = DataLoader(parameters['test_data'], parameters, train_loader)
test_loader(parameters=parameters)
test_data, prune_recall = DocRelationDataset(test_loader, 'test', parameters, train_loader).__call__()
with open(os.path.join(processed_dataset, 'test_test_data.pkl'), 'wb') as f:
pkl.dump(test_data, f, pkl.HIGHEST_PROTOCOL)
with open(os.path.join(processed_dataset, 'test_prune_recall.pkl'), 'wb') as f:
pkl.dump(prune_recall, f, pkl.HIGHEST_PROTOCOL)
m = Trainer(train_loader, parameters, {'train': [], 'test': test_data}, model_folder, prune_recall)
trainer = load_model(parameters['remodelfile'], m)
_, _,_,p,r=trainer.eval_epoch(final=True, save_predictions=True)
print('Saving test metrics ... ', end="")
np.savetxt(parameters['remodelfile']+"/p.txt", p)
np.savetxt(parameters['remodelfile']+"/r.txt", r)
# b = numpy.loadtxt("filename.txt", delimiter=',')
print('DONE')
def main():
print('loading model:')
model=load_model(weight_path)
print('loading dataset:')
test_loader = torch.utils.data.DataLoader(CustomDataset(batch_size,path,'test'),batch_size, shuffle=False, num_workers=1)
print('number of samples',len(test_loader))
test_model(model, test_loader)
def main(config):
if isinstance(config.compared_imgs, str):
imgs = config.compared_imgs.split(' ')
else:
imgs = config.compared_imgs
images = load_and_align_data(config, imgs)
with tf.Graph().as_default():
with tf.Session() as sess:
# Load the model
load_model(config.lfw.valid_model_path)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
phase_train_placeholder = tf.get_default_graph(
).get_tensor_by_name("phase_train:0")
# Run forward pass to calculate embeddings
feed_dict = {
images_placeholder: images,
phase_train_placeholder: False
}
emb = sess.run(embeddings, feed_dict=feed_dict)
nrof_images = len(imgs)
print('Images:')
for i in range(nrof_images):
print('%1d: %s' % (i, imgs[i]))
print('')
# Print distance matrix
print('Distance matrix')
print(' ', end='')
for i in range(nrof_images):
print(' %1d ' % i, end='')
print('')
for i in range(nrof_images):
print('%1d ' % i, end='')
for j in range(nrof_images):
dist = np.sqrt(
np.sum(np.square(np.subtract(emb[i, :], emb[j, :]))))
print(' %1.4f ' % dist, end='')
print('')
def final_evaluation(train_loader, test_loader, valid_loader,
best_model_path_load, model, optimizer, args, dir):
_ = load_model(best_model_path_load, model, optimizer)
model.eval()
exemplars_embedding = load_all_pseudo_input(args, model,
train_loader.dataset)
test_samples = next(iter(test_loader))[0].to(args.device)
visualize_reconstruction(test_samples, model, args, dir)
visualize_generation(train_loader.dataset, model, args, dir)
test_elbo, test_re, test_kl = evaluate_loss(
args,
model,
test_loader,
dataset=train_loader.dataset,
exemplars_embedding=exemplars_embedding)
valid_elbo, valid_re, valid_kl = evaluate_loss(
args,
model,
valid_loader,
dataset=valid_loader.dataset,
exemplars_embedding=exemplars_embedding)
train_elbo, _, _ = evaluate_loss(args,
model,
train_loader,
dataset=train_loader.dataset,
exemplars_embedding=exemplars_embedding)
test_log_likelihood = calculate_likelihood(
args,
model,
test_loader,
exemplars_embedding=exemplars_embedding,
S=args.S)
final_evaluation_txt = 'FINAL EVALUATION ON TEST SET\n' \
'LogL (TEST): {:.2f}\n' \
'LogL (TRAIN): {:.2f}\n' \
'ELBO (TEST): {:.2f}\n' \
'ELBO (TRAIN): {:.2f}\n' \
'ELBO (VALID): {:.2f}\n' \
'RE: {:.2f}\n' \
'KL: {:.2f}'.format(
test_log_likelihood,
0,
test_elbo,
train_elbo,
valid_elbo,
test_re,
test_kl)
print(final_evaluation_txt)
with open(dir + 'vae_experiment_log.txt', 'a') as f:
print(final_evaluation_txt, file=f)
torch.save(test_log_likelihood,
dir + args.model_name + '.test_log_likelihood')
torch.save(test_elbo, dir + args.model_name + '.test_loss')
torch.save(test_re, dir + args.model_name + '.test_re')
torch.save(test_kl, dir + args.model_name + '.test_kl')
def __init__(self, cfg):
self.cfg = cfg
if cfg.arch == 'resnet50':
self.model = get_pose_net(50, 64, cfg.num_classes)
else:
self.model = get_hourglass(cfg.arch, num_classes=cfg.num_classes)
if cfg.pretrained_weights is not None:
weight_file = os.path.join(cfg.save_folder, cfg.pretrained_weights)
load_model(self.model, weight_file)
print("load pretrain mode:{}".format(weight_file))
if cfg.num_gpu > 1 and torch.cuda.is_available():
self.model = torch.nn.DataParallel(self.model).cuda()
else:
self.model = self.model.to(cfg.device)
self.save_folder = cfg.ckpt_dir
self.optim = optim.Adam(self.model.parameters(), lr=cfg.lr)
self.scheduler = optim.lr_scheduler.ExponentialLR(self.optim, gamma=0.99)
def __init__(self):
from utils.utils import load_model
from modeling.TP_Net import Res160, Res320
from modeling.Hourglass import HourglassNet
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if device == 'cpu':
raise EOFError('cpu version for training is not implemented.')
print('Using device: ', device)
self.head = {'center': 1, 'dis': 4, 'line': 1}
self.model = load_model(Res320(self.head), './pretraineds/Res320.pth',
False, False)
self.model = self.model.cuda().eval()
self.in_res = (320, 320)
def get_model(state, args, init_model_name=None):
if init_model_name is not None and os.path.exists(init_model_name):
model, optimizer, state = load_model(init_model_name,
return_optimizer=True,
return_state=True)
else:
if "conv_dropout" in args:
conv_dropout = args.conv_dropout
else:
conv_dropout = cfg.conv_dropout
cnn_args = {1}
if args.fixed_segment is not None:
frames = cfg.frames
else:
frames = None
nb_layers = 4
cnn_kwargs = {
"activation": cfg.activation,
"conv_dropout": conv_dropout,
"batch_norm": cfg.batch_norm,
"kernel_size": nb_layers * [3],
"padding": nb_layers * [1],
"stride": nb_layers * [1],
"nb_filters": [16, 16, 32, 65],
"pooling": [(2, 2), (2, 2), (1, 4), (1, 2)],
"aggregation": args.agg_time,
"norm_out": args.norm_embed,
"frames": frames,
}
nb_frames_staying = cfg.frames // (2**2)
model = CNN(*cnn_args, **cnn_kwargs)
# model.apply(weights_init)
state.update({
'model': {
"name": model.__class__.__name__,
'args': cnn_args,
"kwargs": cnn_kwargs,
'state_dict': model.state_dict()
},
'nb_frames_staying': nb_frames_staying
})
if init_model_name is not None:
save_model(state, init_model_name)
pytorch_total_params = sum(p.numel() for p in model.parameters()
if p.requires_grad)
LOG.info(
"number of parameters in the model: {}".format(pytorch_total_params))
return model, state
def run_cls(config_dir):
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
# ------------------------------------------------------------------------------------------------------------
# 1. classification inference
# ------------------------------------------------------------------------------------------------------------
config_root = Path(config_dir) / 'cls'
config_paths = [config_root / p for p in os.listdir(config_root)]
base_config_paths = [
Path(config_dir) / p for p in os.listdir(config_dir) if 'yml' in p
]
config = load_config(base_config_paths[0])
models = []
for c in config_paths:
models.append(load_model(c))
model = MultiClsModels(models)
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test),
num_classes=config.data.num_classes,
)
all_fnames = []
all_predictions = []
with torch.no_grad():
for i, (batch_fnames, batch_images) in enumerate(tqdm(testloader)):
batch_images = batch_images.to(config.device)
batch_preds = predict_batch(model,
batch_images,
tta=config.test.tta,
task='cls')
all_fnames.extend(batch_fnames)
all_predictions.append(batch_preds)
all_predictions = np.concatenate(all_predictions)
np.save('all_preds', all_predictions)
df = pd.DataFrame(data=all_predictions, index=all_fnames)
df.to_csv('cls_preds.csv')
df.to_csv(KAGGLE_WORK_DIR + '/cls_preds.csv')
def test(data_loader, model, ckp_path, args):
with torch.no_grad():
optimizer = torch.optim.Adagrad(model.parameters())
model = load_model(model, optimizer, ckp_path)
model.eval()
outputs = []
idx_to_name = ['normal', 'smoking', 'calling']
for i, data in tqdm(enumerate(data_loader), total=len(data_loader)):
image = data["image"].cuda()
name = data["name"]
if args.model == "wsdan":
y_pred_raw, _, attention_map = model(image)
with torch.no_grad():
crop_images = batch_augment(image,
attention_map[:, :1, :, :],
mode='crop',
theta=(0.4, 0.6),
padding_ratio=0.1)
y_pred_crop, _, _ = model(crop_images)
output = (y_pred_raw + y_pred_crop) / 2
else:
output = model(image)
output = softmax(output, dim=-1)
for j in range(image.size(0)):
idx = torch.argmax(output[j, :])
category = idx_to_name[idx]
score = output[j, idx]
outputs.append({
"category": category,
"image_name": name[j],
"score": round(float(score), 5)
})
outputs.sort(key=lambda x: int(x['image_name'].split('.')[0]))
with open("./log/result.json", "w+") as f:
json.dump(outputs, f, indent=4)
print("Done.")
return 0
def load_mlp_regressor(model_params, ckpt_path):
"""Load a pre-trained MLP regression model.
Args:
model_params (dict): dictionary of parameters defining the model, for
instance as saved by train_regressor_ood.py in a params.json file
ckpt_path (str): path where the checkpoint is saved.
Returns:
model (regressor.MLP): a pre-trained MLP regression model.
"""
cnn_args = {'dense_sizes':
[int(x) for x in model_params['dense_sizes'].split(',')],
'activation': model_params['activation']}
model = utils.load_model(ckpt_path, regressor.MLP, cnn_args)
return model
def main(config):
data = DataGenerator(config)
with tf.Graph().as_default():
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=config.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
with sess.as_default():
print "loading model ..."
load_model(config.lfw.valid_model_path)
images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")
embedding_size = embeddings.get_shape()[1]
nrof_images = len(data.input)
print('Calculating features for images')
nrof_batches_per_epoch = int(math.ceil(1.0 * nrof_images / config.batch_size))
emb_array = np.zeros((nrof_images, embedding_size))
for i in range(nrof_batches_per_epoch):
start_index = i * config.batch_size
end_index = min((i + 1) * config.batch_size, nrof_images)
paths_batch = data.input[start_index:end_index]
images = load_data(paths_batch, config.lfw.image_size, True) # return 4-d tensor
feed_dict = {images_placeholder: images, phase_train_placeholder: False}
emb_array[start_index:end_index, :] = sess.run(embeddings, feed_dict=feed_dict)
classifier_filename_exp = os.path.expanduser(config.classifier_path)
print('Training classifier')
model = SVC(kernel='linear', probability=True)
model.fit(emb_array, data.y)
print "the compute is -->:", compute_rate(model,config,embeddings,images_placeholder,phase_train_placeholder,embedding_size,sess)
class_names = [cls.name.replace('_', ' ') for cls in data.dataset]
with open(classifier_filename_exp, 'wb') as outfile:
pickle.dump((model, class_names), outfile)
print('Saved classifier model to file "%s"' % classifier_filename_exp)
def bilstm_train_and_eval(train_data,
dev_data,
test_data,
word2id,
tag2id,
crf=True,
remove_O=False,
reload_model=False):
# data
train_word_lists, train_tag_lists = train_data
dev_word_lists, dev_tag_lists = dev_data
test_word_lists, test_tag_lists = test_data
# training
start = time.time()
vocab_size = len(word2id)
out_size = len(tag2id)
# get model_file
if crf:
model_name = "bilstm_crf"
else:
model_name = "bilstm"
emb_size = LSTMConfig.emb_size
hidden_size = LSTMConfig.hidden_size
model_file = "./weights/" + model_name + '_' + str(emb_size) + '_' + str(
hidden_size) + ".pkl"
if reload_model:
# reload trained model!
bilstm_model = load_model(model_file)
else:
# train and save model!
bilstm_model = BILSTM_Model(vocab_size, out_size, crf=crf)
bilstm_model.train(train_word_lists, train_tag_lists, dev_word_lists,
dev_tag_lists, word2id, tag2id)
save_model(
bilstm_model, model_file
) # re-thinking when to save the model? after valid for each epoch?
print("Training finished, taken {} seconds!".format(
int(time.time() - start)))
print("Evaluating {} model:".format(model_name))
pred_tag_lists, test_tag_lists = bilstm_model.test(test_word_lists,
test_tag_lists, word2id,
tag2id)
results_print(test_tag_lists, pred_tag_lists, remove_O=remove_O)
return pred_tag_lists
def main(**kwargs):
options._parse(kwargs)
opt = options
torch.manual_seed(317)
print('Setting up data...')
transforms = T.Compose([T.ToTensor()])
dataset = MotDataset(opt, (640, 480), augment=True, transforms=transforms)
opt = opt.update_dataset_info_and_set_heads(opt, dataset)
print(opt)
os.environ['CUDA_VISIBLE_DEVICES'] = opt.gpus
opt.device = torch.device('cuda' if opt.gpus >= '0' else 'cpu')
print('Creating model...')
model = create_model('dla_34', opt.heads, opt.head_conv)
optimizer = torch.optim.Adam(model.parameters(), opt.lr)
start_epoch = 0
# Get dataloader
train_loader = torch.utils.data.DataLoader(
dataset,
batch_size=opt.batch_size,
shuffle=True,
num_workers=opt.num_workers,
pin_memory=True,
drop_last=True
)
print('Starting training...')
trainer = BaseTrainer(opt, model, optimizer)
trainer.set_device(opt.gpus, -1, opt.device)
if opt.multi_load_model != '':
model, optimizer = load_model(model, opt.multi_load_model, trainer.optimizer)
best = 1e10
for epoch in range(start_epoch + 1, opt.num_epochs + 1):
log_dict_train, _ = trainer.train(epoch, train_loader)
if epoch % opt.save_every == 0:
lr = opt.lr * 0.5
for param_group in optimizer.param_groups:
param_group['lr'] = lr
save_model(os.path.join(opt.save_dir, 'model_{}.pth'.format(epoch)), epoch, model, optimizer)
def make_model(self):
self.model_params = {
'learning_rate': 1e-5,
'kernel_size': 3,
'scale': 2,
'c_dim': 3,
'G0': 25,
'G': 50,
'D': 5,
'C': 3,
}
self.model = load_model(self.model_params,
add_vgg=False,
model_name='rdn',
verbose=False)
self.rdn = self.model.rdn
def get_backbone_layers(weights_path,
nin_name='model_1',
input_layer_name='input_1',
output_layer_name='activation_14'):
backbone = utils.load_model(weights_path, compile=False)
print(backbone.summary())
if nin_name is None:
backbone_in = backbone.get_layer(input_layer_name).input
backbone_out = backbone.get_layer(output_layer_name).output
else:
print(backbone.get_layer(nin_name).summary())
backbone_in = backbone.get_layer(nin_name).get_layer(
input_layer_name).input
backbone_out = backbone.get_layer(nin_name).get_layer(
output_layer_name).output
print(backbone.get_layer(nin_name).summary())
return backbone_in, backbone_out
def get_model_and_loader(config_paths):
config = load_config(config_paths[0])
models = []
for c in config_paths:
models.append(load_model(c))
model = MultiSegModels(models)
testloader = make_loader(
data_folder=config.data.test_dir,
df_path=config.data.sample_submission_path,
phase='test',
img_size=(config.data.height, config.data.width),
batch_size=config.test.batch_size,
num_workers=config.num_workers,
transforms=get_transforms(config.transforms.test))
return model, testloader
def main():
params = get_params()
set_random_seed(params.RANDOM_SEED)
parse_data()
data = DatasetNorm('cutted_data')
train_set, test_set = torch.utils.data.random_split(
data, [data.__len__() - 100, 100])
trainloader = DataLoader(dataset=train_set,
batch_size=params.BATCH_SIZE,
shuffle=True,
num_workers=8)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
tcnn = TempoCNN().to(device)
wandb.init(project="tcnn")
config = wandb.config
config.learning_rate = 0.001
wandb.watch(tcnn)
if not params.LOAD_MODEL:
model = train(tcnn, trainloader)
save_model(model)
else:
model = load_model().to(device)
testloader = DataLoader(dataset=test_set,
batch_size=params.BATCH_SIZE,
shuffle=True)
iters = 0
loss = 0.0
cr_loss = nn.BCELoss()
for i, data in enumerate(testloader, 0):
tcnn.eval()
mels, labels = data[0].to(device), data[1].to(device)
pred = model(mels.unsqueeze(-1).permute(0, 3, 1, 2)).to('cpu').detach()
res = accuracy(pred, labels)
print(res)
loss += cr_loss(pred.float(), labels.float().to('cpu').detach()).item()
iters += 1
print(loss / iters)
def initial_or_load(checkpoint_path_load, model, optimizer, dir):
if os.path.exists(checkpoint_path_load):
model_loaded_str = "******model is loaded*********"
print(model_loaded_str)
with open(dir + 'whole_log.txt', 'a') as f:
print(model_loaded_str, file=f)
checkpoint = load_model(checkpoint_path_load, model, optimizer)
begin_epoch = checkpoint['epoch']
best_loss = checkpoint['best_loss']
e = checkpoint['e']
else:
torch.manual_seed(args.seed)
if args.device == 'cuda':
torch.cuda.manual_seed(args.seed)
random.seed(args.seed)
begin_epoch = 1
best_loss = math.inf
e = 0
return begin_epoch, best_loss, e
def _test(parameters):
model_folder = setup_log(parameters, parameters['save_pred'] + '_test')
print('\nLoading mappings ...')
train_loader = load_mappings(parameters['remodelfile'])
print('\nLoading testing data ...')
test_loader = DataLoader(parameters['test_data'], parameters, train_loader)
test_loader(parameters=parameters)
test_data, prune_recall = DocRelationDataset(test_loader, 'test',
parameters,
train_loader).__call__()
m = Trainer(train_loader, parameters, {
'train': [],
'test': test_data
}, model_folder, prune_recall)
trainer = load_model(parameters['remodelfile'], m)
trainer.eval_epoch(final=True, save_predictions=True)
def load_cnn_classifier(model_params, ckpt_path):
"""Load a pre-trained CNN classifier.
Args:
model_params (dict): dictionary of parameters defining the CNN, for
instance as saved by train_classifier_ood.py in a params.json file
ckpt_path (str): path where the checkpoint is saved.
Returns:
model (classifier.CNN): a pre-trained CNN classifier.
"""
cnn_args = {'conv_dims':
[int(x) for x in model_params['conv_dims'].split(',')],
'conv_sizes':
[int(x) for x in model_params['conv_sizes'].split(',')],
'dense_sizes':
[int(x) for x in model_params['dense_sizes'].split(',')],
'n_classes': model_params['n_classes'], 'onehot': True}
model = utils.load_model(ckpt_path, classifier.CNN, cnn_args)
return model
