def init_vocab(self, data):
assert self.eval == False # for eval vocab must exist
charvocab = CharVocab(data, self.args['shorthand'])
wordvocab = WordVocab(data,
self.args['shorthand'],
cutoff=7,
lower=True)
uposvocab = WordVocab(data, self.args['shorthand'], idx=1)
xposvocab = xpos_vocab_factory(data, self.args['shorthand'])
featsvocab = FeatureVocab(data, self.args['shorthand'], idx=3)
lemmavocab = WordVocab(data,
self.args['shorthand'],
cutoff=7,
idx=4,
lower=True)
deprelvocab = WordVocab(data, self.args['shorthand'], idx=6)
vocab = MultiVocab({
'char': charvocab,
'word': wordvocab,
'upos': uposvocab,
'xpos': xposvocab,
'feats': featsvocab,
'lemma': lemmavocab,
'deprel': deprelvocab
})
return vocab
def get_factory(sh, fn):
print('Resolving vocab option for {}...'.format(sh))
train_file = 'data/pos/{}.train.in.conllu'.format(sh)
if not os.path.exists(train_file):
raise UserWarning(
'Training data for {} not found in the data directory, falling back to using WordVocab. To generate the '
'XPOS vocabulary for this treebank properly, please run the following command first:\n'
'\tstanza/utils/datasets/prepare_pos_treebank.py {}'.format(
fn, fn))
# without the training file, there's not much we can do
key = 'WordVocab(data, shorthand, idx=2)'
return key
doc = CoNLL.conll2doc(input_file=train_file)
data = doc.get([TEXT, UPOS, XPOS, FEATS], as_sentences=True)
print(f'Original length = {len(data)}')
data = filter_data(data, idx=2)
print(f'Filtered length = {len(data)}')
vocab = WordVocab(data, sh, idx=2, ignore=["_"])
key = 'WordVocab(data, shorthand, idx=2, ignore=["_"])'
best_size = len(vocab) - len(VOCAB_PREFIX)
if best_size > 20:
for sep in ['', '-', '+', '|', ',', ':']: # separators
vocab = XPOSVocab(data, sh, idx=2, sep=sep)
length = sum(
len(x) - len(VOCAB_PREFIX) for x in vocab._id2unit.values())
if length < best_size:
key = 'XPOSVocab(data, shorthand, idx=2, sep="{}")'.format(sep)
best_size = length
return key
def xpos_vocab_factory(data, shorthand):
if shorthand in ["af_afribooms", "grc_perseus", "ar_padt", "bg_btb", "hr_set", "cs_cac", "cs_cltt", "cs_fictree", "cs_pdt", "en_partut", "fr_partut", "gl_ctg", "it_isdt", "it_partut", "it_postwita", "it_twittiro", "it_vit", "ja_gsd", "lv_lvtb", "lt_alksnis", "ro_nonstandard", "ro_rrt", "gd_arcosg", "sr_set", "sk_snk", "sl_ssj", "ta_ttb", "uk_iu", "gl_treegal", "la_perseus", "sl_sst"]:
return XPOSVocab(data, shorthand, idx=2, sep="")
elif shorthand in ["en_test", "grc_proiel", "hy_armtdp", "eu_bdt", "be_hse", "ca_ancora", "zh-hant_gsd", "zh-hans_gsdsimp", "lzh_kyoto", "cop_scriptorium", "da_ddt", "en_ewt", "en_gum", "et_edt", "fi_tdt", "fr_ftb", "fr_gsd", "fr_sequoia", "fr_spoken", "de_gsd", "de_hdt", "got_proiel", "el_gdt", "he_htb", "hi_hdtb", "hu_szeged", "ga_idt", "ja_bccwj", "la_proiel", "lt_hse", "mt_mudt", "mr_ufal", "nb_bokmaal", "nn_nynorsk", "nn_nynorsklia", "cu_proiel", "fro_srcmf", "orv_torot", "fa_seraji", "pt_bosque", "pt_gsd", "ru_gsd", "ru_syntagrus", "ru_taiga", "es_ancora", "es_gsd", "swl_sslc", "te_mtg", "tr_imst", "ug_udt", "vi_vtb", "wo_wtb", "bxr_bdt", "et_ewt", "kk_ktb", "kmr_mg", "olo_kkpp", "sme_giella", "hsb_ufal"]:
return WordVocab(data, shorthand, idx=2, ignore=["_"])
elif shorthand in ["nl_alpino", "nl_lassysmall", "la_ittb", "sv_talbanken"]:
return XPOSVocab(data, shorthand, idx=2, sep="|")
elif shorthand in ["en_lines", "sv_lines", "ur_udtb"]:
return XPOSVocab(data, shorthand, idx=2, sep="-")
elif shorthand in ["fi_ftb"]:
return XPOSVocab(data, shorthand, idx=2, sep=",")
elif shorthand in ["id_gsd", "ko_gsd", "ko_kaist"]:
return XPOSVocab(data, shorthand, idx=2, sep="+")
elif shorthand in ["pl_lfg", "pl_pdb"]:
return XPOSVocab(data, shorthand, idx=2, sep=":")
else:
raise NotImplementedError('Language shorthand "{}" not found!'.format(shorthand))
raise UserWarning(
'Training data for {} not found in the data directory, falling back to using WordVocab. To generate the '
'XPOS vocabulary for this treebank properly, please run the following command first:\n'
'\tbash scripts/prep_pos_data.sh {}'.format(fn, fn))
# without the training file, there's not much we can do
key = 'WordVocab(data, shorthand, idx=2)'
mapping[key].append(sh)
continue
doc = Document(
CoNLL.conll2dict(input_file='data/pos/{}.train.in.conllu'.format(sh)))
data = doc.get([TEXT, UPOS, XPOS, FEATS], as_sentences=True)
print(f'Original length = {len(data)}')
data = filter_data(data, idx=2)
print(f'Filtered length = {len(data)}')
vocab = WordVocab(data, sh, idx=2, ignore=["_"])
key = 'WordVocab(data, shorthand, idx=2, ignore=["_"])'
best_size = len(vocab) - len(VOCAB_PREFIX)
if best_size > 20:
for sep in ['', '-', '+', '|', ',', ':']: # separators
vocab = XPOSVocab(data, sh, idx=2, sep=sep)
length = sum(
len(x) - len(VOCAB_PREFIX) for x in vocab._id2unit.values())
if length < best_size:
key = 'XPOSVocab(data, shorthand, idx=2, sep="{}")'.format(sep)
best_size = length
mapping[key].append(sh)
# Generate code. This takes the XPOS vocabulary classes selected above, and generates the
# actual factory class as seen in models.pos.xpos_vocab_factory.
first = True