valid_dataset = moaDataset(fold_valid.sig_id.values,
fold_valid[g_cols].values,
fold_valid[c_cols].values,
fold_valid[cate_cols].values,
fold_valid[score_cols].values,
fold_valid[nonscore_cols].values)
train_loader = DataLoader(train_dataset, batch_size=config.batch_size,
shuffle=True, num_workers=8, drop_last=True)
valid_loader = DataLoader(valid_dataset, batch_size=config.batch_size, shuffle=False, num_workers=8)
model = moaModel(config, num_g, num_c, num_cate,
classes=config.num_scored,
non_classes=config.num_nonscored)
train_module = Learner(model, train_loader, valid_loader, fold, config, seed)
history = train_module.fit(config.n_epochs)
for key in hists.keys():
hists[key].append(history[key])
saved_models['cv'].append(min(history['valid']))
saved_models['path'].append(os.path.join(config.log_dir, f"{config.name}_seed{seed}_fold{fold}.pth"))
# Inference
cv_seed = np.mean([min(hist_valid) for hist_valid in hists['valid']])
print(f'SEED {seed}: CV {cv_seed:.6f}')
cv_score = np.mean(saved_models['cv'])
dataset = moaDataset(test_features.sig_id.values,
test_features[g_cols].values,
test_features[c_cols].values,
test_features[cate_cols].values,
thr_conf = 0.8
alpha = 0.2
if __name__ == '__main__':
seed_everything(config.seed)
model = Unet(encoder_name='resnet50', encoder_weights='imagenet', decoder_use_batchnorm=True,
decoder_attention_type='scse', classes=2, activation=None)
df = pd.read_csv(config.df_path)
train_df = df[df.fold != config.fold].reset_index(drop=True)
valid_df = df[df.fold == config.fold].reset_index(drop=True)
transforms = get_transforms(config.input_size, need=('train', 'val'))
train_dataset = dsbDataset(config.data_dir, config.scr_dir, config.mask_dir, train_df,
tfms=transforms['train'], return_id=False)
valid_dataset = dsbDataset(config.data_dir, config.scr_dir, config.mask_dir, valid_df,
tfms=transforms['val'], return_id=True)
train_loader = DataLoader(dataset=train_dataset, batch_size=config.batch_size, num_workers=config.num_workers,
shuffle=True)
valid_loader = DataLoader(dataset=valid_dataset, batch_size=1, num_workers=config.num_workers,
shuffle=False)
Learner = Learner(model, train_loader, valid_loader, config)
pretrained_path = os.path.join(config.log_dir, 'best_model.pth')
if os.path.isfile(pretrained_path):
Learner.load(pretrained_path)
Learner.log(f"Checkpoint Loaded: {pretrained_path}")
Learner.fit(config.n_epochs)
return correct/len(batch_images)
val_set = torch.utils.data.DataLoader(
torchvision.datasets.MNIST('./', train=False, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize((0.5,), (0.5,))
])),
batch_size=128, shuffle=True)
train_set = torch.utils.data.DataLoader(
torchvision.datasets.MNIST('./', train=True, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize((0.5,), (0.5,))
])),
batch_size=512, shuffle=True)
model = nn.Sequential(nn.Linear(28*28, 2**7),
nn.ReLU(),
nn.Linear(2**7, 2**6),
nn.ReLU(),
nn.Linear(2**6, 10),
nn.LogSoftmax(dim=1))
#optimizer = SgdOptimizer(model.parameters(), 0.003)
optimizer = optim.SGD(model.parameters(), lr=0.003, momentum=0.9)
learner = Learner(train_set, val_set, model, optimizer, nn.NLLLoss(), batch_accuracy)
learner.fit(50)
