72 | from dataclasses import dataclass, field
73 | from typing import Dict, Any
74 | import nlpaug.augmenter.char as nac
75 | import nlpaug.augmenter.word as naw
76 | import nlpaug.augmenter.sentence as nas
77 |
78 | class_translator = {
79 | "ocr": nac.OcrAug,
80 | "contextual_w_e": naw.ContextualWordEmbsAug,
81 | "synonym": naw.SynonymAug,
82 | "backtranslation": naw.BackTranslationAug,
83 | "contextual_s_e": nas.ContextualWordEmbsForSentenceAug,
84 | "abstractive_summ": nas.AbstSummAug,
85 | }
86 |
87 |
88 | [docs]@dataclass
89 | class NLPAugConfig:
90 | """
91 | Configuration for augmenters.
92 |
93 | Parameters
94 | ----------
95 | name : str
96 | Name of the data augmentation technique. Possible values currently are `ocr` (for OCR augmentation), `contextual_w_e`
97 | for Contextual Word Embedding augmentation, `synonym`, `backtranslation`, `contextual_s_e` for Contextual Word Embeddings for Sentence Augmentation,
98 | `abstractive_summ`. If using a custom augmenter class this can be a random name.
99 | augmenter_cls: Any
100 | An optional augmenter class, from `nlpaug` library. Can be used instead of using an identifier name
101 | for loading the class (see param `name` of this class).
102 | proportion : float
103 | Proportion of data augmentation.
104 | aug_kwargs : Dict
105 | Arguments for the data augmentation class. See https://github.com/makcedward/nlpaug/blob/master/example/textual_augmenter.ipynb
106 | """
107 |
108 | name: str = field(metadata={"help": "Name of the data augmentation technique. If using a custom augmenter class this can be a random name."})
109 | augmenter_cls: Any = field(
110 | default=None,
111 | metadata={"help": "An optional augmenter class, from `nlpaug` library. Can be used instead of using an identifier name for loading the class (see param `name` of this class)."}
112 | )
113 | proportion: float = field(
114 | default=0.1, metadata={"help": "proportion of data augmentation"}
115 | )
116 | aug_kwargs: Dict = field(
117 | default=None,
118 | metadata={
119 | "help": "Arguments for the data augmentation class. See https://github.com/makcedward/nlpaug/blob/master/example/textual_augmenter.ipynb"
120 | },
121 | )
122 | ' + 106 | '' + 107 | _("Hide Search Matches") + 108 | "
" 109 | ) 110 | ); 111 | }, 112 | 113 | /** 114 | * helper function to hide the search marks again 115 | */ 116 | hideSearchWords: () => { 117 | document 118 | .querySelectorAll("#searchbox .highlight-link") 119 | .forEach((el) => el.remove()); 120 | document 121 | .querySelectorAll("span.highlighted") 122 | .forEach((el) => el.classList.remove("highlighted")); 123 | localStorage.removeItem("sphinx_highlight_terms") 124 | }, 125 | 126 | initEscapeListener: () => { 127 | // only install a listener if it is really needed 128 | if (!DOCUMENTATION_OPTIONS.ENABLE_SEARCH_SHORTCUTS) return; 129 | 130 | document.addEventListener("keydown", (event) => { 131 | // bail for input elements 132 | if (BLACKLISTED_KEY_CONTROL_ELEMENTS.has(document.activeElement.tagName)) return; 133 | // bail with special keys 134 | if (event.shiftKey || event.altKey || event.ctrlKey || event.metaKey) return; 135 | if (DOCUMENTATION_OPTIONS.ENABLE_SEARCH_SHORTCUTS && (event.key === "Escape")) { 136 | SphinxHighlight.hideSearchWords(); 137 | event.preventDefault(); 138 | } 139 | }); 140 | }, 141 | }; 142 | 143 | _ready(SphinxHighlight.highlightSearchWords); 144 | _ready(SphinxHighlight.initEscapeListener); 145 | -------------------------------------------------------------------------------- /docs/_build/html/_static/twemoji.css: -------------------------------------------------------------------------------- 1 | img.emoji { 2 | height: 1em; 3 | width: 1em; 4 | margin: 0 .05em 0 .1em; 5 | vertical-align: -0.1em; 6 | } 7 | -------------------------------------------------------------------------------- /docs/_build/html/_static/twemoji.js: -------------------------------------------------------------------------------- 1 | function addEvent(element, eventName, fn) { 2 | if (element.addEventListener) 3 | element.addEventListener(eventName, fn, false); 4 | else if (element.attachEvent) 5 | element.attachEvent('on' + eventName, fn); 6 | } 7 | 8 | addEvent(window, 'load', function() { 9 | twemoji.parse(document.body, {'folder': 'svg', 'ext': '.svg'}); 10 | }); 11 | -------------------------------------------------------------------------------- /docs/_build/html/examples.html: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 |In the examples folder you will find example scripts showing how to fine-tune models for different tasks. These tasks are divided in directories, as you see. In all scripts it is also shown how to use ResultsPlotter to save a metrics comparison figure of the models trained.
classificationFor classification we have 2 examples. train_classification.py shows how to train a BERTIN model for multi-class classification (emotion detection, check tweet_eval: emotion dataset for more info.). On the other hand, train_multilabel.py shows how to train a model on a multilabel task.
80 |extractive_qaFor extractive QA we have only 1 example, as this type of task is very similar in all cases: train_sqac.py shows how to train a MarIA-large (Spanish Roberta-large) model on SQAC dataset, with hyperparameter search.
85 |NERFor NER, there is an example script, showing how to train multiple models on multiple NER datasets with different format, where we need to apply a pre_func to one of the datasets. The script is called train_spanish_ner.py.
seq2seqFor this task, check out train_maria_encoder_decoder_marimari.py, which shows how to train a seq2seq model when no encoder-decoder architecture is readily available for a certain language, in this case Spanish. On the other hand, check out train_summarization_mlsum.py to learn how to configure training for two multilingual encoder-decoder models for MLSUM summarization task.
95 |Important: For more detailed tutorials in Jupyter-Notebook format, please check nlpboost notebooks. These tutorials have explanations on all the configuration, which is helpful for getting to better know the tool. They are intended to provide a deep understanding on the different configurations that are needed for each of the tasks, so that the user can easily adapt the scripts for their own tasks and needs.
100 |
76 | Contents:
78 |In this folder you will find examples on how to use nlpboost in a tutorial format, with text explanations for each of the steps needed to configure the experiments. Notebooks can also be run in Google Colab.
The examples are very similar to those in examples folder, in the sense that the datasets chosen for each task are the same. However, in notebooks more models are trained for each dataset, and there are clear explanations on some key aspects of the configuration of each task that the raw scripts lack.
78 || 83 | n | ||
| 87 | |
88 | nlpboost | 89 | |
| 92 | |
93 | nlpboost.augmentation | 94 | |
| 97 | |
98 | nlpboost.augmentation.augmenter_config | 99 | |
| 102 | |
103 | nlpboost.augmentation.TextAugmenterPipeline | 104 | |
| 107 | |
108 | nlpboost.autotrainer | 109 | |
| 112 | |
113 | nlpboost.ckpt_cleaner | 114 | |
| 117 | |
118 | nlpboost.dataset_config | 119 | |
| 122 | |
123 | nlpboost.default_param_spaces | 124 | |
| 127 | |
128 | nlpboost.hfdatasets_manager | 129 | |
| 132 | |
133 | nlpboost.hftransformers_manager | 134 | |
| 137 | |
138 | nlpboost.metrics | 139 | |
| 142 | |
143 | nlpboost.metrics_plotter | 144 | |
| 147 | |
148 | nlpboost.model_config | 149 | |
| 152 | |
153 | nlpboost.results_getter | 154 | |
| 157 | |
158 | nlpboost.tokenization_functions | 159 | |
| 162 | |
163 | nlpboost.utils | 164 | |
