def main_predict(cfg, model_name, verbose=False):
device = get_device()
try:
if model_name == "resnet":
model_path = cfg["models"][model_name]
print(model_path)
model = Resnet(cfg=cfg)
model.load_state_dict(
torch.load(model_path, map_location=torch.device('cpu')))
model.to(device)
print(f'Loaded model from file: {model_path}')
elif model_name == "efficientnet":
model_path = cfg["models"][model_name]
model = EfficientNet(cfg=cfg)
model.load_state_dict(
torch.load(model_path, map_location=torch.device('cpu')))
model.to(device)
print(f'Loaded model from file: {model_path}')
else:
resnet = Resnet(cfg=cfg)
resnet.load_state_dict(
torch.load(cfg["models"]["resnet"], map_location=device))
efficientnet = EfficientNet(cfg=cfg)
efficientnet.load_state_dict(
torch.load(cfg["models"]["efficientnet"],
map_location=torch.device('cpu')))
model = Ensemble(resnet=resnet, efficientnet=efficientnet)
except:
print("Wrong model name")
return
model.eval()
df_test = pd.read_csv(f'{cfg["DATASET_DIR"]}/sample_submission.csv')
test_dataset = DatasetTransformer_prediction(cfg=cfg, df=df_test)
test_dataset_loader = torch.utils.data.DataLoader(
dataset=test_dataset, batch_size=cfg["BATCH_SIZE"], shuffle=True)
if verbose and model_name != "ensemble":
summary(
model,
(3, cfg["IMAGE_SIZE"][model_name], cfg["IMAGE_SIZE"][model_name]))
predictions = get_prediction(model, test_dataset_loader, device,
model_name)
df_test[cfg["TARGET_COLS"]] = predictions
df_test[[cfg["IMAGE_COL"]] + cfg["TARGET_COLS"]].to_csv('submission.csv',
index=False)
if verbose:
print(df_test.head(20))
def get_model(model_name, num_classes):
model = None
if 'resnet' in model_name:
model = Resnet(model_name, num_classes)
elif 'efficientnet' in model_name:
model = EfficientNet(model_name, num_classes)
else:
raise ValueError(f'Undefined model name: {model_name}')
return model
def __init__(self, compound_coef):
super(EfficientNetFeatures, self).__init__()
model = EfficientNet.from_pretrained(f'efficientnet-b{compound_coef}',
False)
del model._conv_head
del model._bn1
del model._avg_pooling
del model._dropout
del model._fc
self.model = model
class EfficientNetMnistTrainer(object):
def __init__(self, input_shape, output_dim):
self.model = EfficientNet(input_shape, output_dim)
optimizer = tf.keras.optimizers.SGD(learning_rate=0.1, momentum=0.1)
self.model.compile(optimizer=optimizer,
loss='categorical_crossentropy',
metrics=['acc'])
def train(
self,
x_train,
t_train,
x_val,
t_val,
epochs,
batch_size,
):
self.history = self.model.fit(x_train,
t_train,
batch_size=batch_size,
epochs=30,
validation_data=(x_val, t_val))
def Efficientnet(model_name, num_classes, test = False):
'''
model_name :'efficientnet-b0', 'efficientnet-b1-7'
'''
model = EfficientNet.from_name(model_name)
if not test:
if LOCAL_PRETRAINED[model_name] == None:
state_dict = load_state_dict_from_url(model_urls[model_name], progress=True)
else:
state_dict = torch.load(LOCAL_PRETRAINED[model_name])
model.load_state_dict(state_dict)
fc_features = model._fc.in_features
model._fc = nn.Linear(fc_features, num_classes)
return model
def model_from_dataset(dataset, **kwargs):
if dataset == 'adult':
return FullyConnected(**kwargs)
elif dataset == 'credit':
return FullyConnected(**kwargs)
elif dataset == 'compass':
return FullyConnected(**kwargs)
elif dataset == 'multi_mnist' or dataset == 'multi_fashion_mnist' or dataset == 'multi_fashion':
return MultiLeNet(**kwargs)
elif dataset == 'celeba':
if 'efficientnet' in kwargs['model_name']:
return EfficientNet.from_pretrained(**kwargs)
elif kwargs['model_name'] == 'resnet18':
return ResNet.from_name(**kwargs)
else:
raise ValueError("Unknown model name {}".format(dataset))
def __init__(self,
input_shape,
encode_dim,
output_dim,
model='efficient_net',
loss='emd'):
self.model = None
if model == 'efficient_net':
self.model = EfficientNet(input_shape, encode_dim, output_dim)
elif model == 'wide_res_net':
self.model = WideResNet(input_shape, output_dim)
else:
raise Exception('no match model name')
optimizer = tf.keras.optimizers.SGD(learning_rate=0.1, momentum=0.1)
loss_func = None
if loss == 'emd':
loss_func = EMD
elif loss == 'categorical_crossentropy':
loss_func = 'categorical_crossentropy'
else:
raise Exception('no match loss function')
self.model.compile(optimizer=optimizer,
loss=loss_func,
metrics=['acc'])
def load_model(config, num_classes):
if config['model']['type'] == 'resnet':
if config['model']['arch'] == 'resnet50':
net = ResNet.resnet50(pretrained=False,
progress=False,
num_classes=num_classes)
elif config['model']['arch'] == 'resnext50':
net = ResNet.resnext50_32x4d(pretrained=False,
progress=False,
num_classes=num_classes)
elif config['model']['arch'] == 'resnet50d':
net = ResNetD.resnet50d(pretrained=False,
progress=False,
num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'tresnet':
if config['model']['arch'] == 'tresnetm':
net = TResNet.TResnetM(num_classes=num_classes)
elif config['model']['arch'] == 'tresnetl':
net = TResNet.TResnetL(num_classes=num_classes)
elif config['model']['arch'] == 'tresnetxl':
net = TResNet.TResnetXL(num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'regnet':
regnet_config = dict()
if config['model']['arch'] == 'regnetx-200mf':
regnet_config['depth'] = 13
regnet_config['w0'] = 24
regnet_config['wa'] = 36.44
regnet_config['wm'] = 2.49
regnet_config['group_w'] = 8
regnet_config['se_on'] = False
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnetx-600mf':
regnet_config['depth'] = 16
regnet_config['w0'] = 48
regnet_config['wa'] = 36.97
regnet_config['wm'] = 2.24
regnet_config['group_w'] = 24
regnet_config['se_on'] = False
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnetx-4.0gf':
regnet_config['depth'] = 23
regnet_config['w0'] = 96
regnet_config['wa'] = 38.65
regnet_config['wm'] = 2.43
regnet_config['group_w'] = 40
regnet_config['se_on'] = False
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnetx-6.4gf':
regnet_config['depth'] = 17
regnet_config['w0'] = 184
regnet_config['wa'] = 60.83
regnet_config['wm'] = 2.07
regnet_config['group_w'] = 56
regnet_config['se_on'] = False
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnety-200mf':
regnet_config['depth'] = 13
regnet_config['w0'] = 24
regnet_config['wa'] = 36.44
regnet_config['wm'] = 2.49
regnet_config['group_w'] = 8
regnet_config['se_on'] = True
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnety-600mf':
regnet_config['depth'] = 15
regnet_config['w0'] = 48
regnet_config['wa'] = 32.54
regnet_config['wm'] = 2.32
regnet_config['group_w'] = 16
regnet_config['se_on'] = True
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnety-4.0gf':
regnet_config['depth'] = 22
regnet_config['w0'] = 96
regnet_config['wa'] = 31.41
regnet_config['wm'] = 2.24
regnet_config['group_w'] = 64
regnet_config['se_on'] = True
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
elif config['model']['arch'] == 'regnety-6.4gf':
regnet_config['depth'] = 25
regnet_config['w0'] = 112
regnet_config['wa'] = 33.22
regnet_config['wm'] = 2.27
regnet_config['group_w'] = 72
regnet_config['se_on'] = True
regnet_config['num_classes'] = num_classes
net = RegNet.RegNet(regnet_config)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'resnest':
if config['model']['arch'] == 'resnest50':
net = ResNest.resnest50(pretrained=False, num_classes=num_classes)
elif config['model']['arch'] == 'resnest101':
net = ResNest.resnest101(pretrained=False, num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'efficient':
if config['model']['arch'] == 'b0':
net = EfficientNet.efficientnet_b0(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b1':
net = EfficientNet.efficientnet_b1(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b2':
net = EfficientNet.efficientnet_b2(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b3':
net = EfficientNet.efficientnet_b3(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b4':
net = EfficientNet.efficientnet_b4(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b5':
net = EfficientNet.efficientnet_b5(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'b6':
net = EfficientNet.efficientnet_b6(pretrained=False,
num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'assembled':
pass
elif config['model']['type'] == 'shufflenet':
if config['model']['arch'] == 'v2_x0_5':
net = ShuffleNetV2.shufflenet_v2_x0_5(pretrained=False,
progress=False,
num_classes=num_classes)
elif config['model']['arch'] == 'v2_x1_0':
net = ShuffleNetV2.shufflenet_v2_x1_0(pretrained=False,
progress=False,
num_classes=num_classes)
elif config['model']['arch'] == 'v2_x1_5':
net = ShuffleNetV2.shufflenet_v2_x1_5(pretrained=False,
progress=False,
num_classes=num_classes)
elif config['model']['arch'] == 'v2_x2_0':
net = ShuffleNetV2.shufflenet_v2_x2_0(pretrained=False,
progress=False,
num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'mobilenet':
if config['model']['arch'] == 'small_075':
net = Mobilenetv3.mobilenetv3_small_075(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'small_100':
net = Mobilenetv3.mobilenetv3_small_100(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'large_075':
net = Mobilenetv3.mobilenetv3_large_075(pretrained=False,
num_classes=num_classes)
elif config['model']['arch'] == 'large_100':
net = Mobilenetv3.mobilenetv3_large_100(pretrained=False,
num_classes=num_classes)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
elif config['model']['type'] == 'rexnet':
if config['model']['arch'] == 'rexnet1.0x':
net = ReXNet.rexnet(num_classes=num_classes, width_multi=1.0)
elif config['model']['arch'] == 'rexnet1.5x':
net = ReXNet.rexnet(num_classes=num_classes, width_multi=1.5)
elif config['model']['arch'] == 'rexnet2.0x':
net = ReXNet.rexnet(num_classes=num_classes, width_multi=2.0)
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['arch']))
else:
raise ValueError('Unsupported architecture: ' +
str(config['model']['type']))
return net
print('loss: %2.3f, top1: %2.3f, top5: %2.3f' %
(tloss / len(loader), top1 / len(loader), top5 / len(loader)))
img_tfs = tfs.Compose([
tfs.Resize((512, 512)),
tfs.ToTensor(),
tfs.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))
])
if __name__ == '__main__':
# net = Classifier()
'''
net = resnet50(True)
'''
net = EfficientNet.from_name('efficientnet-b0')
if os.path.exists(args.model):
net.load_state_dict(torch.load(args.model))
imgnet = ImageFolder(args.data, transform=img_tfs)
loader = DataLoader(imgnet,
args.batch_size,
shuffle=True,
num_workers=6,
pin_memory=True)
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(net.parameters(), args.learning_rate)
#optimizer=torch.optim.SGD(net.parameters(),args.learning_rate,args.momentum,weight_decay=args.weight_decay)
'''
train(loader,net,criterion,optimizer,args.epochs)
'''
validate(loader, net, criterion)
from sklearn.metrics import accuracy_score, roc_auc_score
from tqdm import tqdm
import time
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
opt = TestOptions().parse(print_options=False)
print("{} from {} model testing on {}".format(opt.arch, opt.source_dataset,
opt.target_dataset))
gpu_id = opt.gpu_id
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
use_cuda = torch.cuda.is_available()
print("GPU device %d:" % (gpu_id), use_cuda)
model = EfficientNet.from_name(opt.arch, num_classes=opt.classes)
if opt.resume:
pretrained = opt.resume
print("=> using pre-trained model '{}'".format(pretrained))
model.load_state_dict(torch.load(pretrained)['state_dict'])
model.to('cuda')
cudnn.benchmark = True
print('Total params: %.2fM' % (sum(p.numel()
for p in model.parameters()) / 1000000.0))
criterion = nn.CrossEntropyLoss().cuda()
optimizer = optim.SGD(model.parameters(), lr=opt.lr, momentum=opt.momentum)
def fit(data, fold=None, log=True):
best_score = 0.0
model = EfficientNet("tf_efficientnet_b0_ns").to(device)
# model.load_state_dict(
# torch.load("/content/siim-isic_efficientnet_b0_2.ckpt")[
# "model_state_dict"
# ]
# )
# if log:
# neptune.init("utsav/SIIM-ISIC", api_token=NEPTUNE_API_TOKEN)
# neptune.create_experiment(
# FLAGS["exp_name"],
# exp_description,
# params=FLAGS,
# upload_source_files="*.txt",
# )
optimizer = torch.optim.AdamW(
model.parameters(),
lr=FLAGS["learning_rate"],
weight_decay=FLAGS["weight_decay"],
)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
optimizer,
factor=0.5,
cooldown=0,
mode="min",
patience=3,
verbose=True,
min_lr=1e-8,
)
datasets = get_datasets(data)
# sampler
# labels_vcount = y_train["target"].value_counts()
# class_counts = [
# labels_vcount[0].astype(np.float32),
# labels_vcount[1].astype(np.float32),
# ]
# num_samples = sum(class_counts)
# class_weights = [
# num_samples / class_counts[i] for i in range(len(class_counts))
# ]
# weights = [
# class_weights[y_train["target"].values[i]]
# for i in range(int(num_samples))
# ]
# sampler = WeightedRandomSampler(
# torch.DoubleTensor(weights), int(num_samples)
# )
# loaders
train_loader = DataLoader(
datasets["train"],
batch_size=FLAGS["batch_size"],
num_workers=FLAGS["num_workers"],
shuffle=True, # sampler=sampler,
pin_memory=True,
)
val_loader = DataLoader(
datasets["valid"],
batch_size=FLAGS["batch_size"] * 2,
shuffle=False,
num_workers=FLAGS["num_workers"],
drop_last=True,
)
scaler = GradScaler()
# train loop
for epoch in range(0, FLAGS["num_epochs"]):
print("-" * 27 + f"Epoch #{epoch+1} started" + "-" * 27)
train_loss = train_one_epoch(
train_loader,
model,
optimizer,
epoch,
scheduler=None,
scaler=scaler,
log=log,
)
print(f"\nAverage loss for epoch #{epoch+1} : {train_loss:.5f}")
val_output = val_one_epoch(val_loader, model)
val_loss, auc_score, roc_plot, hist, error_scaled = val_output
scheduler.step(error_scaled)
# logs
# if log:
# neptune.log_metric("AUC/val", auc_score)
# neptune.log_image("ROC/val", roc_plot)
# neptune.log_metric("Loss/val", val_loss)
# neptune.log_image("hist/val", hist)
# checkpoint+upload
if (auc_score > best_score) or (best_score - auc_score < 0.025):
if auc_score > best_score:
best_score = auc_score
save_upload(
model,
optimizer,
best_score,
epoch,
fold,
exp_name=FLAGS["exp_name"],
)
print("-" * 28 + f"Epoch #{epoch+1} ended" + "-" * 28)
# if log:
# neptune.stop()
return model
def initialize_model(model_name,
num_classes,
learning_rate,
use_pretrained=True):
# Initialize these variables which will be set in this if statement. Each of these
# variables is model specific.
model = None
input_size = 0
optimizer = None
criterion = None
if model_name == "resnet101":
""" Resnet101
"""
model = models.resnet101(pretrained=use_pretrained)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "resnet152":
""" Resnet101
"""
model = models.resnet152(pretrained=use_pretrained)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "resnet101-next":
""" Resnet101
"""
model = models.resnext101_32x8d(pretrained=use_pretrained)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "resnet50-next":
""" Resnet101
"""
model = models.resnext50_32x4d(pretrained=use_pretrained)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "wide-resnet101-2":
""" Resnet101
"""
model = models.wide_resnet101_2(pretrained=use_pretrained)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "densenet161":
""" Densenet
"""
model = models.densenet161(pretrained=use_pretrained)
num_ftrs = model.classifier.in_features
model.classifier = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "densenet201":
""" Densenet
"""
model = models.densenet201(pretrained=use_pretrained)
num_ftrs = model.classifier.in_features
model.classifier = nn.Linear(num_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "mobilenet":
""" MobileNet
"""
model = models.mobilenet_v2(pretrained=use_pretrained)
# num_ftrs = model.classifier.in_features
# model.classifier = nn.Linear(num_ftrs, num_classes)
model.classifier[1] = nn.Linear(model.last_channel, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "inception":
""" Inception v3
Be careful, expects (299,299) sized images and has auxiliary output
"""
model = models.inception_v3(pretrained=use_pretrained)
# Handle the auxilary net
num_ftrs = model.AuxLogits.fc.in_features
model.AuxLogits.fc = nn.Linear(num_ftrs, num_classes)
# Handle the primary net
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, num_classes)
input_size = 299
params_to_update = model.parameters()
optimizer = optim.Adam(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
elif model_name == "efficientnet-b3":
model = EfficientNet.from_pretrained(model_name)
in_ftrs = model._fc.in_features
model._fc = nn.Linear(in_ftrs, num_classes)
input_size = 224
params_to_update = model.parameters()
optimizer = optim.RMSprop(params_to_update, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
else:
print("Invalid model name, exiting...")
exit()
return model, optimizer, criterion, input_size
def main_train(cfg, model_name, verbose=False):
device = get_device()
data = get_data(cfg=cfg)
if data is not None:
train_data = data.sample(frac=cfg["TRAIN_VALIDATION_FRAC"])
validation_data = data.drop(train_data.index)
else:
return
train_transform = get_transform(image_size=cfg["IMAGE_SIZE"][model_name],
model_name=model_name,
cfg=cfg,
augmented=True)
validation_transform = get_transform(
image_size=cfg["IMAGE_SIZE"][model_name],
model_name=model_name,
cfg=cfg,
)
train_dataset = DatasetTransformer(train_data,
transform=train_transform,
cfg=cfg)
validation_dataset = DatasetTransformer(validation_data,
transform=validation_transform,
cfg=cfg)
train_dataset_loader = torch.utils.data.DataLoader(
dataset=train_dataset, batch_size=cfg["BATCH_SIZE"], shuffle=True)
validation_dataset_loader = torch.utils.data.DataLoader(
dataset=validation_dataset, batch_size=cfg["BATCH_SIZE"], shuffle=True)
if model_name == "resnet":
model = Resnet(cfg=cfg, pretrained=True)
pass
elif model_name == "efficientnet":
model = EfficientNet(cfg=cfg, pretrained=True)
else:
print("Wrong model name")
return
model.to(device)
if verbose:
summary(
model,
(3, cfg["IMAGE_SIZE"][model_name], cfg["IMAGE_SIZE"][model_name]))
optimizer = torch.optim.Adam(model.parameters(),
lr=cfg["LR"],
weight_decay=cfg["WEIGHT_DECAY"])
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer=optimizer,
patience=3,
factor=0.5,
verbose=verbose,
mode='min',
threshold=2 * 1e-2)
loss = nn.BCELoss()
train_updater = GraphUpdater(type="Train")
validation_updater = GraphUpdater(type="Validation")
model_checkpoint = ModelCheckpoint(model)
print(f"Training on {len(train_dataset)} images")
for t in range(cfg["EPOCHS"]):
print(f'Epoch: {t}')
train_loss, train_auc = train(model=model,
data_loader=train_dataset_loader,
loss_function=loss,
optimizer=optimizer,
device=device)
train_updater.update(**{"loss": train_loss, "accuracy": train_auc})
print(f'\tTraining step: Loss: {train_loss}, AUC: {train_auc}',
end='\n')
val_loss, val_auc = test(model=model,
data_loader=validation_dataset_loader,
loss_function=loss,
device=device)
validation_updater.update(**{"loss": val_loss, "accuracy": val_auc})
print(f'\tValidation step: Loss: {val_loss}, AUC: {val_auc}', end='\n')
torch.save(model.state_dict(), f'checkpoint_epoch_{t}.pth')
print('\tModel saved')
model_checkpoint.update(loss=val_loss)
scheduler.step(val_loss)
train_updater.display()
validation_updater.display()
from utils.aug import data_augment, rand_bbox
from utils.train_utils import save_checkpoint, adjust_learning_rate
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
from options.base import BaseOptions
opt = BaseOptions().parse(print_options=False)
#print("{} from {} model testing on {}".format(opt.arch, opt.source_dataset, opt.target_dataset))
gpu_id = opt.gpu_id
os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_id)
use_cuda = torch.cuda.is_available()
print("GPU device %d:" %(gpu_id), use_cuda)
model = EfficientNet.from_name(opt.arch, num_classes=opt.classes,
override_params={'dropout_rate': opt.dropout, 'drop_connect_rate': opt.dropconnect})
model.to('cuda')
cudnn.benchmark = True
best_acc = 0
data_dir = opt.source_dataset
train_dir = os.path.join(data_dir, 'train')
train_aug = transforms.Compose([
transforms.Lambda(lambda img: data_augment(img, opt)),
transforms.Resize(opt.size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
# transforms.RandomErasing(p=0.3, scale=(0.02, 0.10), ratio=(0.3, 3.3), value=0, inplace=True),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
FLAGS["num_workers"] = 8
FLAGS["learning_rate"] = 4e-4
FLAGS["num_epochs"] = 40
FLAGS["weight_decay"] = 1e-4
FLAGS["log_steps"] = 20
FLAGS["img_size"] = IMG_SIZE
FLAGS["loss"] = "focal"
FLAGS["optimizer"] = "AdamW"
FLAGS["scheduler"] = "ReduceLROnPlateau"
FLAGS["exp_name"] = "enet_b0"
FLAGS["fold"] = [0] # , 1, 2, 3, 4]
FLAGS["val_freq"] = 1
FLAGS["num_cores"] = 8
FLAGS["seed"] = 42
model_cpu = EfficientNet("tf_efficientnet_b0_ns")
WRAPPED_MODEL = xmp.MpModelWrapper(model_cpu)
SERIAL_EXEC = xmp.MpSerialExecutor()
for fold_no in FLAGS["fold"]:
X_train = df_train[df_train["fold"] != fold_no][[
col for col in df_train.columns if col != "target"
]]
X_val = df_train[df_train["fold"] == fold_no][[
col for col in df_train.columns if col != "target"
]]
y_train = df_train[df_train["fold"] != fold_no][[
col for col in df_train.columns if col == "target"
]]
y_val = df_train[df_train["fold"] == fold_no][[
col for col in df_train.columns if col == "target"
def __init__(self, input_shape, output_dim):
self.model = EfficientNet(input_shape, output_dim)
optimizer = tf.keras.optimizers.SGD(learning_rate=0.1, momentum=0.1)
self.model.compile(optimizer=optimizer,
loss='categorical_crossentropy',
metrics=['acc'])