def create_parser():
parser = yaap.Yaap()
# data options
parser.add_str("dataset", default="woz")
parser.add_pth("exp-dir",
is_dir=True,
must_exist=True,
required=True,
help="Path to an experiment folder.")
# model options
parser.add_int("gpu",
min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# embedding evaluation options
parser.add_str("embed-type",
default="glove",
choices=("glove", "bin", "hdf5"),
help="Type of embedding to load for evaluation.")
parser.add_pth("embed-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"tests/data/glove/"
"glove.840B.300d.woz.txt")),
help="Path to embedding file for evaluation.")
# input/output options
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
# inference options
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap()
# data options
parser.add_str("dataset", default="woz")
parser.add_pth("exp-dir", is_dir=True, must_exist=True, required=True,
help="Path to an experiment folder.")
parser.add_str("dirname", default="z-interp",
help="Name of the resulting directory.")
parser.add_str("anchor1", regex=r"(train|dev|test)-\d+",
help="Data index of the first anchor. If not provided, "
"a random data point will be chosen.")
parser.add_str("anchor2", regex=r"(train|dev|test)-\d+",
help="Data index of the second anchor. If not provided, "
"a random data point will be chosen.")
# interpolation options
parser.add_int("steps", default=10,
help="Number of intermediate steps between two data points.")
# model options
parser.add_int("gpu", min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# display options
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
return parser
def create_parser():
parser = yaap.Yaap()
# data options
parser.add_pth("data-path", must_exist=True,
help="Path to the data. If not given, then the data "
"will be generated from the model's prior.")
parser.add_pth("processor-path", must_exist=True, required=True,
help="Path to the processor pickle file.")
# model options
parser.add_pth("model-path", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/vhda-mini.yml")),
help="Path to the model configuration file.")
parser.add_pth("ckpt-path", must_exist=True, required=True,
help="Path to the model checkpoint.")
# model-specific options (TDA)
parser.add_flt("conv-scale", default=1.0,
help="Scale to introduce into conv vector "
"for TDA generation.")
parser.add_flt("spkr-scale", default=1.0,
help="Scale to introduce into spkr vector "
"for TDA generation.")
parser.add_flt("goal-scale", default=1.0,
help="Scale to introduce into goal vector "
"for TDA generation.")
parser.add_flt("state-scale", default=1.0,
help="Scale to introduce into state vector "
"for TDA generation.")
parser.add_flt("sent-scale", default=1.0,
help="Scale to introduce into sent vector "
"for TDA generation.")
# model-specific options (general)
parser.add_int("beam-size", default=4,
help="Beam search beam size.")
parser.add_int("max-sent-len", default=30,
help="Beam search maximum sentence length.")
# generation options
parser.add_int("batch-size", default=32,
help="Mini-batch size.")
parser.add_bol("validate-dst",
help="Whether to validate generated samples "
"to be a valid dst dialogs.")
parser.add_bol("validate-unique",
help="Whether to validate by checking uniqueness.")
parser.add_int("instances",
help="Number of dialog instances to generate. "
"If not given, the same number of instances "
"as the data will be generated.")
# misc options
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir", default=pathlib.Path("out"),
help="Directory to save output generation files.")
parser.add_int("gpu", min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap(
desc="Create z-interpolation between two random data points")
# data options
parser.add_pth("data-dir",
is_dir=True,
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"tests/data/json")),
help="Path to the data dir. Must contain 'train.json' and "
"'dev.json'.")
parser.add_str("splits",
is_list=True,
default=("train", ),
choices=("train", "dev", "test"),
help="List of splits to evaluate on.")
parser.add_pth("processor-path",
required=True,
must_exist=True,
help="Path to the processor pickle file.")
parser.add_str("anchor1",
regex=r"(train|dev|test)-\d+",
help="Data index of the first anchor. If not provided, "
"a random data point will be chosen.")
parser.add_str("anchor2",
regex=r"(train|dev|test)-\d+",
help="Data index of the second anchor. If not provided, "
"a random data point will be chosen.")
# interpolation options
parser.add_int(
"steps",
default=10,
help="Number of intermediate steps between two data points.")
# model options
parser.add_pth("model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/vhda-mini.yml")),
help="Path to the model configuration file.")
parser.add_pth("ckpt-path",
required=True,
must_exist=True,
help="Path to the model checkpoint.")
parser.add_int("gpu",
min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# display options
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir",
default="out",
is_dir=True,
help="Directory to save output files.")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
return parser
def create_parser():
parser = yaap.Yaap()
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("../../tests/config/logging.yml")),
help="Path to a logging configuration file.")
parser.add_pth("data-dir", is_dir=True, must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("../../tests/data/json")),
help="Path to a json-format dialogue dataset.")
parser.add_pth("save-dir", is_dir=True, default="out",
help="Saving directory.")
parser.add_bol("overwrite", help="Whether to overwrite.")
parser.add_pth("model-path", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/glad-mini.yml")))
parser.add_int("epochs", default=10, help="Number of epochs")
parser.add_int("batch-size", default=32, help="Batch size.")
parser.add_bol("scheduled-lr", help="Enable scheduled learning rate.")
parser.add_str("scheduler-cls", default="StepLR",
help="Name of the scheduler class under "
"`torch.optim.lr_scheduler` package.")
parser.add_str("scheduler-kwargs",
default="{\"step\": 10, \"gamma\": 0.8}",
help="Keyword arguments for the scheduler class, given "
"as a serialized json dictionary.")
parser.add_str("loss", default="sum", choices=("sum", "mean"),
help="Type of loss aggregation ('sum' or 'mean').")
parser.add_flt("l2norm",
help="Weight of l2norm regularization.")
parser.add_flt("gradient-clip",
help="Clipping bounds for gradients.")
parser.add_bol("train-validate",
help="Whether to validate on the training set as well.")
parser.add_bol("early-stop", help="Whether to early stop.")
parser.add_str("early-stop-criterion", default="joint-goal",
help="Early stopping criterion.")
parser.add_int("early-stop-patience",
help="Number of epochs to wait until early stopped.")
parser.add_bol("validate-asr",
help="Whether to use asr information during validation.")
parser.add_bol("test-asr",
help="Whether to use asr information during testing.")
parser.add_str("asr-method", default="scaled",
choices=("score", "uniform", "ones", "scaled"),
help="Type of aggregation method to use when summing output "
"scores during asr-enabled evaluation.")
parser.add_str("asr-sigmoid-sum-order", default="sigmoid-sum",
help="The order of sum and sigmoid operations in ASR mode.")
parser.add_int("asr-topk", min_bound=1,
help="Number of top-k candidates.")
parser.add_bol("save-ckpt",
help="Whether to save the final checkpoint. "
"If enabled, it will be saved as 'ckpt.pth'"
"under save_dir.")
parser.add_int("gpu", help="gpu device id.")
return parser
def create_parser():
parser = yaap.Yaap("Conduct generative data augmentation experiments "
"(with easier arguments).")
# data options
parser.add_str("dataset", default="woz")
parser.add_pth("exp-dir",
is_dir=True,
must_exist=True,
required=True,
help="Path to an experiment folder.")
# model options
parser.add_str("dst-model",
default="gce-ft-wed",
help="Name of the DST model.")
# model-specific options (TDA)
parser.add_flt("scale",
is_list=True,
num_elements=5,
default=(1.0, 1.0, 1.0, 1.0, 1.0),
help="Scale to introduce into conv, goal and sent vector "
"for TDA generation.")
# generation options
parser.add_int("gen-runs", default=3, help="Number of generation trials")
parser.add_flt("multiplier",
default=1.0,
help="Ratio of dialog instances to generate. ")
parser.add_bol("validate-unique",
help="Whether to validate by checking uniqueness.")
# DST options
parser.add_int("batch-size", default=100)
parser.add_int("dst-runs",
default=5,
help="Number of DST models to train and evaluate using "
"different seeds.")
parser.add_int("epochs",
default=200,
help="Number of epochs to train DST. "
"The actual number of epochs will be scaled by "
"the multiplier.")
parser.add_bol("test-asr",
help="Whether to use asr information during testing.")
# misc options
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir",
default=pathlib.Path("out"),
help="Directory to save output generation files.")
parser.add_int("gpu",
min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap()
parser.add_pth("data-path", must_exist=True, required=True,
help="Path to the dialog data.")
parser.add_str("data-format", default="json",
choices=("woz", "json", "dstc"),
help="Data format of the data to be loaded.")
parser.add_int("max-column-length", default=50,
help="Maximum length of each column of formatted table.")
parser.add_str("save-format", default="human",
choices=("human", "woz", "json"),
help="Output data format.")
parser.add_pth("save-path",
help="Path to save the resulting text file.")
return parser
def create_parser():
parser = yaap.Yaap()
# data options
parser.add_pth("data-dir", is_dir=True, must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("tests/data/json")),
help="Path to the data dir. Must contain 'train.json' and "
"'dev.json'.")
parser.add_str("eval-splits", is_list=True,
default=("train", "dev", "test"),
choices=("train", "dev", "test"),
help="List of splits to evaluate on.")
parser.add_pth("processor-path", required=True, must_exist=True,
help="Path to the processor pickle file.")
# model options
parser.add_pth("model-path", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/vhda-mini.yml")),
help="Path to the model configuration file.")
parser.add_pth("ckpt-path", required=True, must_exist=True,
help="Path to the model checkpoint.")
parser.add_int("gpu", min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# embedding evaluation options
parser.add_str("embed-type", default="glove",
choices=("glove", "bin", "hdf5"),
help="Type of embedding to load for evaluation.")
parser.add_pth("embed-path", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("tests/data/glove/"
"glove.840B.300d.woz.txt")),
help="Path to embedding file for evaluation.")
# input/output options
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir", default="out",
help="Path to save evaluation results.")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
# inference options
parser.add_int("batch-size", default=32,
help="Mini-batch size.")
parser.add_int("beam-size", default=4)
parser.add_int("max-conv-len", default=20)
parser.add_int("max-sent-len", default=30)
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap()
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"../../tests/config/logging.yml")),
help="Path to a logging configuration file.")
parser.add_pth("save-path",
default="out.json",
help="Path to save evaluation results (json).")
parser.add_pth("data-dir",
is_dir=True,
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"../../tests/data/json")),
help="Path to a json-format dialogue dataset.")
parser.add_pth("model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/glad-mini.yml")))
parser.add_pth("ckpt-path",
must_exist=True,
required=True,
help="Path to a checkpoint file.")
parser.add_pth("processor-path",
must_exist=True,
required=True,
help="Path to a processor object.")
parser.add_int("batch-size", default=32, help="Batch size.")
parser.add_str("loss",
default="sum",
choices=("sum", "mean"),
help="Type of loss aggregation ('sum' or 'mean').")
parser.add_bol("test-asr",
help="Whether to use asr information during testing.")
parser.add_str(
"asr-method",
default="scaled",
choices=("score", "uniform", "ones", "scaled"),
help="Type of aggregation method to use when summing output "
"scores during asr-enabled evaluation.")
parser.add_str("asr-sigmoid-sum-order",
default="sigmoid-sum",
help="The order of sum and sigmoid operations in ASR mode.")
parser.add_int("asr-topk", min_bound=1, help="Number of top-k candidates.")
parser.add_int("gpu", help="gpu device id.")
return parser
def create_parser():
parser = yaap.Yaap(
desc="Conducts t-test on a summary or a pair of summary files.")
parser.add_pth("summary-path", required=True, must_exist=True)
parser.add_pth("other-path",
must_exist=True,
help="If supplied, the script will perform unpaired "
"t-test of two individual samples on common keys.")
parser.add_str("expected-mean",
is_list=True,
regex=r"(.+)=(.+)",
help="Expected mean for summary samples. Must be "
"given as a pair of key and expected population "
"mean (e.g. 'loss=0.3'). Could supply multiple "
"pairs. Applicable only when no other summary "
"path is supplied.")
parser.add_pth("save-path")
return parser
def create_parser():
parser = yaap.Yaap()
parser.add_pth("model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"examples/model-vhda.yml")),
help="Path to a base model path.")
parser.add_pth("run-path",
must_exist=True,
required=True,
help="Path to a base run configuration path.")
parser.add_str("storage",
format="url",
default="sqlite:///examples/study.db",
help="Optuna database url supported by sqlalchemy.")
parser.add_str("study-name", default="default", help="Optuna study name.")
parser.add_int("num-trials", help="Number of trials.")
parser.add_int("num-jobs", default=1, help="Number of concurrent jobs.")
parser.add_flt("timeout", help="Timeout for a single trial in seconds.")
return parser
def create_parser():
parser = yaap.Yaap()
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).parent
.joinpath("../examples/logging.yml")),
help="Path to logging configuration file.")
parser.add_pth("data-dir", is_dir=True, must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("examples")),
help="Path to the data dir. Must contain 'train.json', "
"'valid.json' and 'test.json'.")
parser.add_str("data-format", default="json",
choices=("woz", "json", "dstc"),
help="Data format of the data to be loaded.")
parser.add_pth("glad-dir", is_dir=True, must_exist=True,
default=(pathlib.Path(__file__).parent
.joinpath("../dst/glad").absolute()),
help="Directory to an existing glad codebase.")
parser.add_pth("save-dir", is_dir=True, default="out-glad",
help="Directory for saving output files.")
parser.add_int("max-epochs", min_bound=1, default=50,
help="Maximum epochs to train models.")
parser.add_int("batch-size", min_bound=1, default=50,
help="Mini-batch size during stochastic gd.")
parser.add_flt("emb-dropout", default=0.2,
help="Embedding dropout.")
parser.add_flt("local-dropout", default=0.2,
help="Local dropout.")
parser.add_flt("global-dropout", default=0.2,
help="Global dropout.")
parser.add_str("early-stop-criterion", default="joint_goal",
choices=("joint_goal", "turn_inform",
"turn_request", "hmean"))
parser.add_int("runs", default=10,
help="Number of runs to execute and aggregate.")
parser.add_int("gpu", help="Index of specific GPU device to use.")
return parser
def create_parser():
parser = yaap.Yaap()
# data options
parser.add_pth("data-dir",
is_dir=True,
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"tests/data/json")),
help="Path to the data dir. Must contain 'train.json' and "
"'dev.json'.")
parser.add_str("eval-splits",
default=("train", "dev", "test"),
choices=("train", "dev", "test"),
help="List of splits to evaluate on.")
# model options
parser.add_pth("gen-model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/vhda-mini.yml")),
help="Path to the generative model configuration file.")
parser.add_pth("dst-model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"dst/internal/configs/gce.yml")),
help="Path to the dst model configuration file.")
parser.add_int("gpu",
min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# model-specific options (TDA)
parser.add_flt("conv-scale",
default=1.0,
help="Scale to introduce into conv vector "
"for TDA generation.")
parser.add_flt("spkr-scale",
default=1.0,
help="Scale to introduce into spkr vector "
"for TDA generation.")
parser.add_flt("goal-scale",
default=1.0,
help="Scale to introduce into goal vector "
"for TDA generation.")
parser.add_flt("state-scale",
default=1.0,
help="Scale to introduce into state vector "
"for TDA generation.")
parser.add_flt("sent-scale",
default=1.0,
help="Scale to introduce into sent vector "
"for TDA generation.")
# display options
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir",
default="out",
is_dir=True,
help="Directory to save output files.")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
parser.add_int("report-every",
help="Report training statistics every N steps.")
# training options
parser.add_int("batch-size", default=32, help="Mini-batch size.")
parser.add_str("optimizer",
default="adam",
choices=("adam", ),
help="Optimizer to use.")
parser.add_flt("gradient-clip", help="Clip gradients by norm size.")
parser.add_flt("l2norm-weight", help="Weight of l2-norm regularization.")
parser.add_flt("learning-rate",
default=0.001,
min_bound=0,
help="Optimizer learning rate.")
parser.add_int("epochs",
default=10,
min_bound=1,
help="Number of epochs to train for.")
parser.add_str("kld-schedule",
help="KLD w schedule given as a list of data points. Each "
"data point is a pair of training step and target "
"dropout scale. Steps in-between data points will be "
"interpolated. e.g. '[(0, 1.0), (10000, 0.1)]'")
parser.add_str(
"dropout-schedule",
help="Dropout schedule given as a list of data points. Each "
"data point is a pair of training step and target "
"dropout scale. Steps in-between data points will be "
"interpolated. e.g. '[(0, 1.0), (10000, 0.1)]'")
parser.add_bol("disable-kl", help="Whether to disable kl-divergence term.")
parser.add_str("kl-mode",
default="kl-mi",
help="KL mode: one of kl, kl-mi, kl-mi+.")
# validation options
parser.add_int("valid-batch-size",
default=32,
help="Mini-batch sizes for validation inference.")
parser.add_flt("validate-every",
default=1,
help="Number of epochs in-between validations.")
parser.add_bol("early-stop", help="Whether to enable early-stopping.")
parser.add_str("early-stop-criterion",
default="~loss",
help="The training statistics key to use as criterion "
"for early-stopping. Prefix with '~' to denote "
"negation during comparison.")
parser.add_int("early-stop-patience",
help="Number of epochs to wait without breaking "
"records until executing early-stopping.")
parser.add_int("beam-size", default=4)
parser.add_int("max-conv-len", default=20)
parser.add_int("max-sent-len", default=30)
# testing options
parser.add_str("embed-type",
default="glove",
choices=("glove", "bin", "hdf5"),
help="Type of embedding to load for emb. evaluation.")
parser.add_pth("embed-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"tests/data/glove/"
"glove.840B.300d.woz.txt")),
help="Path to embedding file for emb. evaluation.")
# generation options
parser.add_int("gen-runs", default=3, help="Number of generation runs.")
parser.add_bol("validate-dst",
help="Whether to validate by checking uniqueness.")
parser.add_bol("validate-unique",
help="Whether to validate by checking uniqueness.")
parser.add_flt("multiplier",
default=1.0,
help="Ratio of dialog instances to generate. ")
# DST options
parser.add_int("dst-batch-size", default=32, help="Mini-batch size.")
parser.add_int("dst-runs",
default=5,
help="Number of DST models to train and evaluate using "
"different seeds.")
parser.add_int("dst-epochs",
default=200,
help="Number of epochs to train DST. "
"The actual number of epochs will be scaled by "
"the multiplier.")
parser.add_flt("dst-l2norm", help="DST Weight of l2norm regularization.")
parser.add_flt("dst-gradient-clip",
help="DST Clipping bounds for gradients.")
parser.add_bol("dst-test-asr",
help="Whether to use asr information during testing.")
# misc
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap("Conduct generative data augmentation experiments.")
# data options
parser.add_pth("data-dir",
is_dir=True,
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"tests/data/json")),
help="Path to a json-format dialogue dataset.")
parser.add_pth("processor-path",
must_exist=True,
required=True,
help="Path to the processor pickle file.")
# model options
parser.add_pth("gen-model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/vhda-mini.yml")),
help="Path to the generative model configuration file.")
parser.add_pth("dst-model-path",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"dst/internal/configs/gce.yml")),
help="Path to the dst model configuration file.")
parser.add_pth("ckpt-path",
must_exist=True,
required=True,
help="Path to the model checkpoint.")
# model-specific options (TDA)
parser.add_flt("conv-scale",
default=1.0,
help="Scale to introduce into conv vector "
"for TDA generation.")
parser.add_flt("spkr-scale",
default=1.0,
help="Scale to introduce into spkr vector "
"for TDA generation.")
parser.add_flt("goal-scale",
default=1.0,
help="Scale to introduce into goal vector "
"for TDA generation.")
parser.add_flt("state-scale",
default=1.0,
help="Scale to introduce into state vector "
"for TDA generation.")
parser.add_flt("sent-scale",
default=1.0,
help="Scale to introduce into sent vector "
"for TDA generation.")
# model-specific options (general)
parser.add_int("beam-size", default=4, help="Beam search beam size.")
parser.add_int("max-sent-len",
default=30,
help="Beam search maximum sentence length.")
# generation options
parser.add_int("gen-runs", default=3, help="Number of generations to run.")
parser.add_int("gen-batch-size", default=32, help="Mini-batch size.")
parser.add_flt("multiplier",
default=1.0,
help="Ratio of dialog instances to generate. ")
parser.add_bol("validate-unique",
help="Whether to validate by checking uniqueness.")
# DST options
parser.add_int("dst-batch-size", default=32, help="Mini-batch size.")
parser.add_int("dst-runs",
default=5,
help="Number of DST models to train and evaluate using "
"different seeds.")
parser.add_int("epochs",
default=200,
help="Number of epochs to train DST. "
"The actual number of epochs will be scaled by "
"the multiplier.")
parser.add_flt("l2norm", help="DST Weight of l2norm regularization.")
parser.add_flt("gradient-clip", help="DST Clipping bounds for gradients.")
parser.add_bol("test-asr",
help="Whether to use asr information during testing.")
# misc options
parser.add_pth("logging-config",
must_exist=True,
default=(pathlib.Path(__file__).absolute().parent.joinpath(
"configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir",
default=pathlib.Path("out"),
help="Directory to save output generation files.")
parser.add_int("gpu",
min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
parser.add_int("seed", help="Random seed.")
return parser
def create_parser():
parser = yaap.Yaap()
# data options
parser.add_pth("data-dir", is_dir=True, must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("tests/data/json")),
help="Path to the data dir. Must contain 'train.json' and "
"'dev.json'.")
# model options
parser.add_pth("model-path", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/vhda-mini.yml")),
help="Path to the model configuration file.")
parser.add_int("gpu", min_bound=0,
help="GPU device to use. (e.g. 0, 1, etc.)")
# display options
parser.add_pth("logging-config", must_exist=True,
default=(pathlib.Path(__file__).absolute().parent
.joinpath("configs/logging.yml")),
help="Path to a logging config file (yaml/json).")
parser.add_pth("save-dir", default="out", is_dir=True,
help="Directory to save output files.")
parser.add_bol("overwrite", help="Whether to overwrite save dir.")
parser.add_int("report-every",
help="Report training statistics every N steps.")
# training options
parser.add_int("batch-size", default=32,
help="Mini-batch size.")
parser.add_int("valid-batch-size", default=32,
help="Mini-batch sizes for validation inference.")
parser.add_str("optimizer", default="adam", choices=("adam",),
help="Optimizer to use.")
parser.add_flt("gradient-clip",
help="Clip gradients by norm size.")
parser.add_flt("l2norm-weight",
help="Weight of l2-norm regularization.")
parser.add_flt("learning-rate", default=0.001, min_bound=0,
help="Optimizer learning rate.")
parser.add_int("epochs", default=10, min_bound=1,
help="Number of epochs to train for.")
parser.add_str("kld-schedule",
help="KLD w schedule given as a list of data points. Each "
"data point is a pair of training step and target "
"dropout scale. Steps in-between data points will be "
"interpolated. e.g. '[(0, 1.0), (10000, 0.1)]'")
parser.add_str("dropout-schedule",
help="Dropout schedule given as a list of data points. Each "
"data point is a pair of training step and target "
"dropout scale. Steps in-between data points will be "
"interpolated. e.g. '[(0, 1.0), (10000, 0.1)]'")
parser.add_flt("validate-every", default=1,
help="Number of epochs in-between validations.")
parser.add_bol("early-stop",
help="Whether to enable early-stopping.")
parser.add_str("early-stop-criterion", default="~loss",
help="The training statistics key to use as criterion "
"for early-stopping. Prefix with '~' to denote "
"negation during comparison.")
parser.add_int("early-stop-patience",
help="Number of epochs to wait without breaking "
"records until executing early-stopping. "
"defaults to infinity.")
parser.add_int("save-every",
help="Number of epochs to wait until saving a model "
"checkpoint.")
# model specific settings
parser.add_bol("disable-kl",
help="Whether to disable kl-divergence term.")
parser.add_str("kl-mode", default="kl-mi",
help="KL mode: one of kl, kl-mi, kl-mi+.")
parser.add_int("seed", help="Random seed.")
return parser