def main(): parser = argparse.ArgumentParser() parser.add_argument("--src", help=f"Path of source file to read from.", required=True) parser.add_argument("--hyp", help="Path of file to write hypothesis to.", required=True) parser.add_argument("--mod", help="Path of model file to read.", required=True) args = parser.parse_args() exp_dir = os.path.dirname(__file__) exp = "{EXP}" param_collection.ParamManager.init_param_col() # TODO: can we avoid the LoadSerialized proxy and load stuff directly? load_experiment = persistence.LoadSerialized(filename=args.mod) uninitialized_experiment = persistence.YamlPreloader.preload_obj( load_experiment, exp_dir=exp_dir, exp_name=exp) loaded_experiment = persistence.initialize_if_needed( uninitialized_experiment) model = loaded_experiment.model inference = model.inference param_collection.ParamManager.populate() decoding_task = eval_task.DecodingEvalTask(args.src, args.hyp, model, inference) decoding_task.eval()
def setUp(self): events.clear() ParamManager.init_param_col() # Load a pre-trained model load_experiment = LoadSerialized(filename=f"test/data/tiny_jaen.model", overwrite=[ { "path": "train", "val": None }, { "path": "status", "val": None }, ]) EXP_DIR = '.' EXP = "decode" uninitialized_experiment = YamlPreloader.preload_obj(load_experiment, exp_dir=EXP_DIR, exp_name=EXP) loaded_experiment = initialize_if_needed(uninitialized_experiment) ParamManager.populate() # Pull out the parts we need from the experiment self.model = loaded_experiment.model src_vocab = self.model.src_reader.vocab trg_vocab = self.model.trg_reader.vocab event_trigger.set_train(False) self.src_data = list( self.model.src_reader.read_sents("test/data/head.ja")) self.trg_data = list( self.model.trg_reader.read_sents("test/data/head.en"))
def test_shallow(self): test_obj = yaml.load(""" a: !DummyArgClass arg1: !DummyArgClass2 _xnmt_id: id1 v: some_val arg2: !Ref { name: id1 } """) preloaded = persistence.YamlPreloader.preload_obj(root=test_obj, exp_name="exp1", exp_dir=self.out_dir) initalized = persistence.initialize_if_needed(preloaded) persistence.save_to_file(self.model_file, initalized)
def initialize_generator(self, **kwargs): # TODO: refactor? if kwargs.get("len_norm_type", None) is None: len_norm = xnmt.length_normalization.NoNormalization() else: len_norm = initialize_if_needed(kwargs["len_norm_type"]) search_args = {} if kwargs.get("max_len", None) is not None: search_args["max_len"] = kwargs["max_len"] if kwargs.get("beam", None) is None: self.search_strategy = GreedySearch(**search_args) else: search_args["beam_size"] = kwargs.get("beam", 1) search_args["len_norm"] = len_norm self.search_strategy = BeamSearch(**search_args) self.report_path = kwargs.get("report_path", None) self.report_type = kwargs.get("report_type", None)
def test_simple_reference(self): test_obj = yaml.load(""" !DummyArgClass arg1: !DummyArgClass arg1: !DummyArgClass2 { v: some_val } arg2: !DummyArgClass2 { v: some_other_val } arg2: !Ref { path: arg1 } """) preloaded = persistence.YamlPreloader.preload_obj(root=test_obj, exp_name="exp1", exp_dir=self.out_dir) initialized = persistence.initialize_if_needed(preloaded) dump = persistence._dump(initialized) reloaded = yaml.load(dump) if isinstance(reloaded.arg1, persistence.Ref): reloaded.arg1, reloaded.arg2 = reloaded.arg2, reloaded.arg1 self.assertIsInstance(reloaded.arg1, DummyArgClass) self.assertIsInstance(reloaded.arg2, persistence.Ref) self.assertIsInstance(reloaded.arg1.arg1, DummyArgClass2) self.assertIsInstance(reloaded.arg1.arg2, DummyArgClass2)
def main(overwrite_args=None): with tee.Tee(), tee.Tee(error=True): argparser = argparse.ArgumentParser() argparser.add_argument("--dynet-mem", type=str) argparser.add_argument("--dynet-seed", type=int, help="set random seed for DyNet and XNMT.") argparser.add_argument("--dynet-autobatch", type=int) argparser.add_argument("--dynet-devices", type=str) argparser.add_argument("--dynet-viz", action='store_true', help="use visualization") argparser.add_argument("--dynet-gpu", action='store_true', help="use GPU acceleration") argparser.add_argument("--dynet-gpu-ids", type=int) argparser.add_argument("--dynet-gpus", type=int) argparser.add_argument("--dynet-weight-decay", type=float) argparser.add_argument("--dynet-profiling", type=int) argparser.add_argument("--settings", type=str, default="standard", help="settings (standard, debug, or unittest)" "must be given in '=' syntax, e.g." " --settings=standard") argparser.add_argument("experiments_file") argparser.add_argument("experiment_name", nargs='*', help="Run only the specified experiments") argparser.set_defaults(generate_doc=False) args = argparser.parse_args(overwrite_args) if args.dynet_seed: random.seed(args.dynet_seed) np.random.seed(args.dynet_seed) if args.dynet_gpu: if settings.CHECK_VALIDITY: settings.CHECK_VALIDITY = False log_preamble( "disabling CHECK_VALIDITY because it is not supported on GPU currently", logging.WARNING) config_experiment_names = YamlPreloader.experiment_names_from_file( args.experiments_file) results = [] # Check ahead of time that all experiments exist, to avoid bad surprises experiment_names = args.experiment_name or config_experiment_names if args.experiment_name: nonexistent = set(experiment_names).difference( config_experiment_names) if len(nonexistent) != 0: raise Exception("Experiments {} do not exist".format(",".join( list(nonexistent)))) log_preamble( f"running XNMT revision {tee.get_git_revision()} on {socket.gethostname()} on {datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}" ) for experiment_name in experiment_names: ParamManager.init_param_col() uninitialized_exp_args = YamlPreloader.preload_experiment_from_file( args.experiments_file, experiment_name) logger.info(f"=> Running {experiment_name}") glob_args = uninitialized_exp_args.data.exp_global log_file = glob_args.log_file if os.path.isfile(log_file) and not settings.OVERWRITE_LOG: logger.warning( f"log file {log_file} already exists, skipping experiment; please delete log file by hand if you want to overwrite it " f"(or activate OVERWRITE_LOG, by either specifying an environment variable as OVERWRITE_LOG=1, " f"or specifying --settings=debug, or changing xnmt.settings.Standard.OVERWRITE_LOG manually)" ) continue tee.set_out_file(log_file) model_file = glob_args.model_file uninitialized_exp_args.data.exp_global.commandline_args = args # Create the model experiment = initialize_if_needed(uninitialized_exp_args) ParamManager.param_col.model_file = experiment.exp_global.model_file ParamManager.param_col.save_num_checkpoints = experiment.exp_global.save_num_checkpoints ParamManager.populate() # Run the experiment eval_scores = experiment(save_fct=lambda: save_to_file( model_file, experiment, ParamManager.param_col)) results.append((experiment_name, eval_scores)) print_results(results) tee.unset_out_file()
def main(overwrite_args: Optional[Sequence[str]] = None) -> None: with tee.Tee(), tee.Tee(error=True): argparser = argparse.ArgumentParser() utils.add_backend_argparse(argparser) argparser.add_argument("--settings", type=str, default="standard", help="settings (standard, debug, or unittest)" "must be given in '=' syntax, e.g." " --settings=standard") argparser.add_argument( "--resume", action='store_true', help="whether a saved experiment is being resumed, and" "locations of output files should be re-used.") argparser.add_argument("--backend", type=str, default="dynet", help="backend (dynet or torch)") argparser.add_argument("experiments_file") argparser.add_argument("experiment_name", nargs='*', help="Run only the specified experiments") argparser.set_defaults(generate_doc=False) args = argparser.parse_args(overwrite_args) if xnmt.backend_dynet and args.dynet_seed: args.seed = args.dynet_seed if getattr(args, "seed", None): random.seed(args.seed) np.random.seed(args.seed) if xnmt.backend_torch: torch.manual_seed(0) if xnmt.backend_dynet and args.dynet_gpu and settings.CHECK_VALIDITY: settings.CHECK_VALIDITY = False log_preamble( "disabling CHECK_VALIDITY because it is not supported in the DyNet/GPU setting", logging.WARNING) config_experiment_names = YamlPreloader.experiment_names_from_file( args.experiments_file) results = [] # Check ahead of time that all experiments exist, to avoid bad surprises experiment_names = args.experiment_name or config_experiment_names if args.experiment_name: nonexistent = set(experiment_names).difference( config_experiment_names) if len(nonexistent) != 0: raise Exception("Experiments {} do not exist".format(",".join( list(nonexistent)))) log_preamble( f"running XNMT revision {tee.get_git_revision()} on {socket.gethostname()} with {'DyNet' if xnmt.backend_dynet else 'PyTorch'} on {datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}" ) for experiment_name in experiment_names: ParamManager.init_param_col() uninitialized_exp_args = YamlPreloader.preload_experiment_from_file( args.experiments_file, experiment_name, resume=args.resume) logger.info(f"=> Running {experiment_name}") glob_args = uninitialized_exp_args.data.exp_global log_file = glob_args.log_file if not settings.OVERWRITE_LOG: log_files_exist = [] if os.path.isfile(log_file): log_files_exist.append(log_file) if os.path.isdir(log_file + ".tb"): log_files_exist.append(log_file + ".tb/") if log_files_exist: logger.warning( f"log file(s) {' '.join(log_files_exist)} already exists, skipping experiment; " f"please delete log file by hand if you want to overwrite it " f"(or activate OVERWRITE_LOG, by either specifying an environment variable OVERWRITE_LOG=1, " f"or specifying --settings=debug, or changing xnmt.settings.Standard.OVERWRITE_LOG manually)" ) continue elif settings.OVERWRITE_LOG and os.path.isdir(log_file + ".tb"): shutil.rmtree( log_file + ".tb/" ) # remove tensorboard logs from previous run that is being overwritten tee.set_out_file(log_file, exp_name=experiment_name) try: model_file = glob_args.model_file uninitialized_exp_args.data.exp_global.commandline_args = vars( args) # Create the model experiment = initialize_if_needed(uninitialized_exp_args) ParamManager.param_col.model_file = experiment.exp_global.model_file ParamManager.param_col.save_num_checkpoints = experiment.exp_global.save_num_checkpoints ParamManager.populate() # Run the experiment eval_scores = experiment( save_fct=lambda: save_to_file(model_file, experiment)) results.append((experiment_name, eval_scores)) print_results(results) except Exception as e: file_logger.error(traceback.format_exc()) raise e finally: tee.unset_out_file()
import xnmt.tee from xnmt.param_collection import ParamManager from xnmt.persistence import initialize_if_needed, YamlPreloader, LoadSerialized, save_to_file EXP_DIR = os.path.dirname(__file__) EXP = "programmatic-load" model_file = f"{EXP_DIR}/models/{EXP}.mod" log_file = f"{EXP_DIR}/logs/{EXP}.log" xnmt.tee.set_out_file(log_file) ParamManager.init_param_col() load_experiment = LoadSerialized( filename=f"{EXP_DIR}/models/programmatic.mod", overwrite=[ {"path" : "train", "val" : None} ] ) uninitialized_experiment = YamlPreloader.preload_obj(load_experiment, exp_dir=EXP_DIR, exp_name=EXP) loaded_experiment = initialize_if_needed(uninitialized_experiment) # if we were to continue training, we would need to set a save model file like this: # ParamManager.param_col.model_file = model_file ParamManager.populate() # run experiment loaded_experiment(save_fct=lambda: save_to_file(model_file, loaded_experiment))