sparse_caption.utils package

Submodules

sparse_caption.utils.config module

Created on Thu Jun 22 14:48:09 2017

@author: jiahuei

class sparse_caption.utils.config.Config(x: Optional[str] = None, **kwargs)

Bases: object

Configuration object.

check_loaded_version(loaded_version)
compat()
deepcopy()
dict()
get(key, default_value)
json(**kwargs)
classmethod load_config_json(config_filepath, verbose=True)
save_config(exist_ok=True)

Save this instance to a json file.

Parameters

exist_ok (bool) – If set to True, allow overwrites.

update(kv_mapping)

sparse_caption.utils.file module

Utilities for file download and caching.

sparse_caption.utils.file.dump_json(path, data, utf8=True, lf_newline=True, **json_kwargs)
sparse_caption.utils.file.dumps_file(path, string, utf8=True, lf_newline=True)
sparse_caption.utils.file.file_size(path, suffix='B')
sparse_caption.utils.file.get_file(fname, origin, dest_dir, md5_hash=None, file_hash=None, hash_algorithm='auto', extract=False, archive_format='auto')

Downloads a file from a URL if it not already in the cache.

The file at the url origin is downloaded to the dest_dir, and given the filename fname. The final location of a file example.txt would therefore be dest_dir/example.txt.

Files in tar, tar.gz, tar.bz, and zip formats can also be extracted. Passing a hash will verify the file after download. The command line programs shasum and sha256sum can compute the hash.

Parameters

  • fname – Name of the file. If an absolute path /path/to/file.txt is specified the file will be saved at that location.

  • origin – Original URL of the file.

  • dest_dir – Location to store the files.

  • md5_hash – Deprecated in favor of ‘file_hash’. md5 hash of the file for verification

  • file_hash – The expected hash string of the file after download. The sha256 and md5 hash algorithms are both supported.

  • hash_algorithm – Select the hash algorithm to verify the file. options are ‘md5’, ‘sha256’, and ‘auto’. The default ‘auto’ detects the hash algorithm in use.

  • extract – True tries extracting the file as an Archive, like tar or zip.

  • archive_format – Archive format to try for extracting the file. Options are ‘auto’, ‘tar’, ‘zip’, and None. ‘tar’ includes tar, tar.gz, and tar.bz files. The default ‘auto’ is [‘tar’, ‘zip’]. None or an empty list will return no matches found.

Returns

Path to the downloaded file

sparse_caption.utils.file.list_dir(path)
sparse_caption.utils.file.list_files(path)
sparse_caption.utils.file.load_pil_image_from_url(url)
sparse_caption.utils.file.read_file(path)
sparse_caption.utils.file.read_json(path, utf8=True)
sparse_caption.utils.file.tqdm_hook(t)

Wraps tqdm instance. Don’t forget to close() or __exit__() the tqdm instance once you’re done with it (easiest using with syntax). .. rubric:: Example

with tqdm(…) as t:

reporthook = tqdm_hook(t) urllib.urlretrieve(…, reporthook=reporthook)

sparse_caption.utils.file.validate_file(fpath, file_hash, algorithm='auto', chunk_size=65535)

Validates a file against a sha256 or md5 hash.

Parameters

  • fpath – path to the file being validated

  • file_hash – The expected hash string of the file. The sha256 and md5 hash algorithms are both supported.

  • algorithm – Hash algorithm, one of ‘auto’, ‘sha256’, or ‘md5’. The default ‘auto’ detects the hash algorithm in use.

  • chunk_size – Bytes to read at a time, important for large files.

Returns

Whether the file is valid

sparse_caption.utils.file.zip_dir(target_dir, save_dir)

sparse_caption.utils.losses module

Created on 16 Sep 2020 15:00:29 @author: jiahuei

class sparse_caption.utils.losses.LabelSmoothing(size=0, padding_idx=0, smoothing=0.0)

Bases: torch.nn.modules.module.Module

Implement label smoothing.

forward(input, target, mask)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class sparse_caption.utils.losses.LanguageModelCriterion

Bases: torch.nn.modules.module.Module

forward(input, target, mask)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

training: bool
class sparse_caption.utils.losses.RewardCriterion

Bases: torch.nn.modules.module.Module

forward(input, mask, reward)
This function computes

log(y_t) * reward * mask_t (where mask_t zeroes out non-words in the sequence)

given

input = predicted probability sequence = predicted word index reward = …

training: bool

sparse_caption.utils.misc module

Created on 29 Apr 2020 15:55:21 @author: jiahuei

Utility functions.

class sparse_caption.utils.misc.ChoiceList(choices)

Bases: object

A Type for ArgParse for validation of choices. https://mail.python.org/pipermail/tutor/2011-April/082825.html

sparse_caption.utils.misc.configure_logging(logging_level: Union[int, str] = 20, logging_fmt: str = '%(levelname)s: %(name)s: %(funcName)s: %(message)s', logger_obj: Union[None, logging.Logger] = None) → logging.Logger

Setup logging on the root logger, because transformers calls logger.info upon import.

Adapted from:

https://stackoverflow.com/a/54366471/5825811 Configures a simple console logger with the given level. A use-case is to change the formatting of the default handler of the root logger.

Format variables:

https://docs.python.org/3/library/logging.html#logrecord-attributes

sparse_caption.utils.misc.csv_to_float_list(input_string: str)
sparse_caption.utils.misc.csv_to_int_list(input_string: str)
sparse_caption.utils.misc.csv_to_str_list(input_string: str)
sparse_caption.utils.misc.get_memory_info()

Get node total memory and memory usage https://stackoverflow.com/a/17718729

sparse_caption.utils.misc.grouper(iterable, group_n, fill_value=0)
sparse_caption.utils.misc.humanise_number(size: Union[float, int], suffix: str = 'B')
sparse_caption.utils.misc.replace_from_right(string: str, old: str, new: str, count: int = - 1)

String replacement from right to left. https://stackoverflow.com/a/3679215

sparse_caption.utils.misc.str_to_bool(input_string: str)
sparse_caption.utils.misc.str_to_none(input_string: str)
sparse_caption.utils.misc.str_to_sequence(input_string: str)
sparse_caption.utils.misc.time_function(func)

sparse_caption.utils.model_utils module

@author: jiahuei, ruotianluo

sparse_caption.utils.model_utils.clones(module, N)

Produce N identical layers.

sparse_caption.utils.model_utils.count_nonzero(tensor)
sparse_caption.utils.model_utils.densify_state_dict(state_dict)
sparse_caption.utils.model_utils.length_average(length, logprobs, alpha=0.0)

Returns the average probability of tokens in a sequence.

sparse_caption.utils.model_utils.length_wu(length, logprobs, alpha=0.0)

NMT length re-ranking score from “Google’s Neural Machine Translation System” :cite:`wu2016google`.

sparse_caption.utils.model_utils.map_recursive(x: Any, func: Callable)

Applies func to elements of x recursively. :param x: An item or a potentially nested structure of tuple, list or dict. :param func: A single argument function.

Returns

The same x but with func applied.

sparse_caption.utils.model_utils.map_to_cuda(x: Any)
sparse_caption.utils.model_utils.pack_wrapper(module, att_feats, att_masks)
sparse_caption.utils.model_utils.pad_unsort_packed_sequence(inputs, inv_ix)
sparse_caption.utils.model_utils.penalty_builder(penalty_config)
sparse_caption.utils.model_utils.reorder_beam(tensor: torch.Tensor, beam_idx: torch.Tensor, beam_dim: int = 0)
sparse_caption.utils.model_utils.repeat_tensors(n, x, dim=0)

For a tensor of size Bx…, we repeat it n times, and make it Bnx… For collections, do nested repeat

sparse_caption.utils.model_utils.requires_grad(model, flag=True)
sparse_caption.utils.model_utils.sequence_from_numpy(sequence)
sparse_caption.utils.model_utils.set_seed(seed: int)
sparse_caption.utils.model_utils.sort_pack_padded_sequence(inputs, lengths)
sparse_caption.utils.model_utils.split_tensors(n, x)
sparse_caption.utils.model_utils.to_cuda(x: Any)

sparse_caption.utils.natural_sort module

Created on Thu Jun 28 15:36:44 2018

@author: jiahuei

sparse_caption.utils.natural_sort.atoi(text)
sparse_caption.utils.natural_sort.natural_keys(text)

alist.sort(key=natural_keys) sorts in human order http://nedbatchelder.com/blog/200712/human_sorting.html (See Toothy’s implementation in the comments)

sparse_caption.utils.optim module

Created on 16 Sep 2020 15:01:35 @author: jiahuei

class sparse_caption.utils.optim.CosineOpt(optimizer, max_train_step, learning_rate_init, learning_rate_min)

Bases: sparse_caption.utils.optim.RateOpt

Optim wrapper that implements rate.

rate()

Implement lrate above

class sparse_caption.utils.optim.NoamOpt(optimizer, model_size, factor, warmup)

Bases: sparse_caption.utils.optim.RateOpt

Optim wrapper that implements rate.

rate()

Implement lrate above

class sparse_caption.utils.optim.RateOpt

Bases: object

Optim wrapper that implements rate.

step(step=None, epoch=None)

Update parameters and rate

class sparse_caption.utils.optim.StepLROpt(optimizer, learning_rate_init, learning_rate_decay_start, learning_rate_decay_every, learning_rate_decay_rate)

Bases: sparse_caption.utils.optim.RateOpt

Optim wrapper that implements rate.

rate()

Implement lrate above

sparse_caption.utils.optim.build_optimizer(params, config)
sparse_caption.utils.optim.clip_gradient(optimizer, grad_clip)
sparse_caption.utils.optim.get_optim(parameters, config)

sparse_caption.utils.training module

Created on 01 May 2020 15:00:54 @author: jiahuei

https://github.com/huggingface/transformers/blob/v2.9.0/examples/lightning_base.py

class sparse_caption.utils.training.TrainingModule(config: sparse_caption.utils.config.Config)

Bases: object

Base class for training and evaluation.

ALL_METRICS = ['Bleu_1', 'Bleu_2', 'Bleu_3', 'Bleu_4', 'METEOR', 'ROUGE_L', 'CIDEr', 'SPICE']
SCST_BASELINE = ['greedy', 'sample']
SCST_SAMPLE = ['beam_search', 'random']
static add_argparse_args(parser: Union[argparse._ArgumentGroup, argparse.ArgumentParser])
checkpoint_path

alias of str

collate_fn: Dict[str, Callable]
compute_scst_loss(model_inputs, gts, loss_fn)
config: sparse_caption.utils.config.Config
data: sparse_caption.data.karpathy.KarpathyDataset
classmethod eval_model(state_dict, config, split='test')
eval_on_split(loader, split)
get_dataloader(split: str, collate_fn: Callable, generation_mode: bool = False)
maybe_load_checkpoint(strict=True)
model: torch.nn.modules.module.Module
optimizer: torch.optim.optimizer.Optimizer
prepare()
scst_scorer: sparse_caption.scst.scorers.CaptionScorer
test_dataloader()
tokenizer: sparse_caption.tokenizer.Tokenizer
train_dataloader()
val_dataloader()

Module contents

Created on 01 Aug 2020 18:12:28 @author: jiahuei