def main(): train_dpath = os.path.join(sys.argv[1], "train") # no / ending valid_dpath = os.path.join(sys.argv[1], "validation") # no / endin model_fpath = sys.argv[2] train_dataset = class_model.Dataset(train_dpath, mode="train") valid_dataset = class_model.Dataset(valid_dpath, mode="validation") model = AE(img_size=IMG_SIZE).cuda() model = _train(model, train_dataset, valid_dataset) torch.save(model.state_dict(), model_fpath)
def main(): root_dpath = sys.argv[1] model_fpath = sys.argv[2] train_dataset = class_model.Dataset(os.path.join(root_dpath, "train"), "train") valid_dataset = class_model.Dataset(os.path.join(root_dpath, "validation"), "train") train_valid_dataset = ConcatDataset([train_dataset, valid_dataset]) model = AE_chu().cuda() model = train(train_valid_dataset, model) torch.save(model.state_dict(), model_fpath) return
def main(): train_dpath = os.path.join(sys.argv[1], "train") # no / ending valid_dpath = os.path.join(sys.argv[1], "validation") # no / endin model_fpath = sys.argv[2] train_dataset = class_model.Dataset(train_dpath, mode="train") valid_dataset = class_model.Dataset(valid_dpath, mode="validation") model = class_model.kCNN().cuda() model = _train(model, train_dataset, valid_dataset) # fine tune with validation train_valid_dataset = ConcatDataset([train_dataset, valid_dataset]) model = _train(model, train_valid_dataset, valid_dataset=None) #save model torch.save(model.state_dict(), model_fpath) return
def main(): root_dpath = sys.argv[1] model_fpath = sys.argv[2] train_dataset = class_model.Dataset(os.path.join(root_dpath, "train"), mode="test") valid_dataset = class_model.Dataset(os.path.join(root_dpath, "validation"), mode="test") model = AE().cuda() model.load_state_dict(torch.load(model_fpath)) model.eval() train_code = get_encoded_vector(train_dataset, model) valid_code = get_encoded_vector(valid_dataset, model) predict_acc(root_dpath, train_code, valid_code) return
def main(): test_dpath = sys.argv[1] model_path = sys.argv[2] output_path = sys.argv[3] test_dataset = class_model.Dataset(test_dpath, mode="test") model = class_model.kCNN().cuda() model.load_state_dict(torch.load(model_path)) #test ans_list = test(model, test_dataset) #submit.csv with open(output_path, 'w') as f: f.write('image_id,label\n') for i in range(len(ans_list)): f.write('{},{}\n'.format(test_dataset.filename_dict[i], chr(ans_list[i]+ord('A')))) return