if opt.layers != -1: opt.enc_layers = opt.layers opt.dec_layers = opt.layers opt.brnn = (opt.encoder_type == "brnn") opt.pre_word_vecs = os.path.join(opt.embd, 'embedding') print(vars(opt)) json.dump(opt.__dict__, open(os.path.join( opt.save_path, 'opt.json'), 'w'), sort_keys=True, indent=2) #if torch.cuda.is_available(): torch.cuda.set_device(0) device = torch.device("cuda") set_seed(opt.seed) # Set up the logging server. # logger = Logger(os.path.join(opt.save_path, 'tb')) def report_func(epoch, batch, num_batches, start_time, lr, report_stats): """ This is the user-defined batch-level traing progress report function. Args: epoch(int): current epoch count. batch(int): current batch count. num_batches(int): total number of batches.
# arg_parser.add_argument('-tgt_vocab', help="Path to an existing target vocabulary") # arg_parser.add_argument('-report_every', type=int, default=100000, help="Report status every this many sentences") # --- options.set_common_options(arg_parser) options.set_preprocess_options(arg_parser) args = arg_parser.parse_args() args.train_anno = os.path.join(args.root_dir, args.dataset, 'train.json') args.valid_anno = os.path.join(args.root_dir, args.dataset, 'dev.json') args.test_anno = os.path.join(args.root_dir, args.dataset, 'test.json') args.save_data = os.path.join(args.root_dir, args.dataset) if args.cuda and args.seed is not None: set_seed(args.seed) def main(): fields = table.IO.TableDataset.get_fields() logger.info(" * building training") train = table.IO.TableDataset(args.train_anno, fields, args.permute_order, args, True) if os.path.isfile(args.valid_anno): logger.info(" * building valid") valid = table.IO.TableDataset(args.valid_anno, fields, permute_order=0, args=args,