def setUp(self):
self.xc1 = xc_load(os.path.join(testfile_dir, 'xigt/kor-ex.xml'))
self.inst = self.xc1[0]
self.g1_2 = xigt_find(self.inst, id='g1.2')
self.m2_1 = xigt_find(self.inst, id="m2.1")
self.p1 = xigt_find(self.inst, id='p1')
self.w1 = xigt_find(self.inst, id="w1")
self.m_tier = xigt_find(self.inst, id="m")
self.new_m = Item(id="m5", tier=self.m_tier, segmentation="w1[0:2]+w4[2:3]")
def evaluate_instance(inst, classifier, tagger):
# Get the supervised POS tags...
"""
:param inst:
:type inst: RGIgt
:param classifier: MalletMaxent
:param tagger: StanfordPOSTagger
"""
sup_gloss_tier = pos_tag_tier(inst, GLOSS_WORD_ID) # We will incrementally build up the tag sequences...
sup_lang_tier = pos_tag_tier(inst, LANG_WORD_ID)
sup_tags = []
prj_tags = []
cls_tags = []
# If there are no supervised tags on the gloss line, but there are on the language line...
if sup_gloss_tier is None and sup_lang_tier is not None:
try:
add_gloss_lang_alignments(inst)
project_lang_to_gloss(inst)
sup_gloss_tier = pos_tag_tier(inst, GLOSS_WORD_ID)
except RGXigtException:
pass
if sup_gloss_tier:
# Do the classification
classify_gloss_pos(inst, classifier)
cls_tier = pos_tag_tier(inst, GLOSS_WORD_ID, tag_method=INTENT_POS_CLASS)
for sup_item in sup_gloss_tier:
word = xigt_find(inst, id=sup_item.alignment)
if not word:
continue
else:
word = word.value()
# prj_item = xigt_find(prj_tier, alignment=sup_item.alignment)
# if prj_item is None:
# prj_tag = 'UNK'
# else:
# prj_tag = prj_item.value()
cls_item = xigt_find(cls_tier, alignment=sup_item.alignment)
if cls_item is None:
cls_tag = 'UNK'
else:
cls_tag = cls_item.value()
sup_tags.append(POSToken(word, label=sup_item.value()))
# prj_tags.append(POSToken(word, label=prj_tag))
cls_tags.append(POSToken(word, label=cls_tag))
return sup_tags, cls_tags
def evaluate_classifier_on_instances(inst_list, classifier, feat_list, pos_class_matrix, gold_tagmap=None):
"""
Given a list of instances, do the evaluation on them.
:param inst_list:
:param classifier:
:param tagger:
:return:
"""
pd = load_posdict() if (CLASS_FEATS_DICT in feat_list) or (CLASS_FEATS_PDICT in feat_list) or (CLASS_FEATS_NDICT in feat_list) else False
matches = 0
compares = 0
for inst in inst_list:
sup_postier = gloss_tag_tier(inst, tag_method=INTENT_POS_MANUAL)
if sup_postier is None:
continue
gw_tier = gloss(inst)
classify_gloss_pos(inst, classifier,
posdict=pd,
feat_prev_gram=CLASS_FEATS_PRESW in feat_list,
feat_next_gram=CLASS_FEATS_NEXSW in feat_list,
feat_dict=CLASS_FEATS_DICT in feat_list,
feat_prev_gram_dict=CLASS_FEATS_PDICT in feat_list,
feat_next_gram_dict=CLASS_FEATS_NDICT in feat_list,
feat_suffix=CLASS_FEATS_SUF in feat_list,
feat_prefix=CLASS_FEATS_PRE in feat_list,
feat_morph_num=CLASS_FEATS_NUMSW in feat_list,
feat_has_number=CLASS_FEATS_NUM in feat_list,
feat_basic=CLASS_FEATS_SW in feat_list)
cls_postier = gloss_tag_tier(inst, tag_method=INTENT_POS_CLASS)
for cls_tag in cls_postier:
word = xigt_find(gw_tier, id=cls_tag.alignment)
sup_tag = xigt_find(sup_postier, alignment=cls_tag.alignment)
if sup_tag is None:
continue
else:
sup_tag_v = sup_tag.value()
if gold_tagmap is not None:
sup_tag_v = gold_tagmap.get(sup_tag_v)
pos_class_matrix.add(sup_tag_v, cls_tag.value())
if cls_tag.value() == sup_tag_v:
matches += 1
compares += 1
return matches, compares, matches/compares*100
def word_align_test(self):
"""
Test that the gloss has been automatically aligned at the word level correctly.
"""
at = Alignment()
for gw in gloss(self.igt):
gw_idx = item_index(gw)
lw = xigt_find(self.igt, id=gw.alignment)
if lw is not None:
at.add((gw_idx, item_index(lw)))
self.assertEqual(at, Alignment([(1,1),(2,2),(3,3),(4,4)]))
def extract_tagger_from_instance(inst: Igt, output_stream, pos_source, tm):
"""
Given an instance, retrieve the language-line words and POS tags.
:param inst:
:param output_stream:
:param pos_source:
"""
lang_pos_tags = lang_tag_tier(inst, tag_method=pos_source)
lang_words = lang(inst)
training_sentences = 0
# -------------------------------------------
# Only try extracting if there are in fact valid POS tags.
# -------------------------------------------
if lang_pos_tags:
first = True
for lang_word in lang_words:
lang_pos_tag = None
if lang_pos_tags is not None:
lang_pos_tag = xigt_find(lang_pos_tags, alignment=lang_word.id)
tag_string = lang_pos_tag.value() if lang_pos_tag is not None else handle_unknown_pos(inst, lang_word)
if tag_string and tm:
tag_string = tm[tag_string]
word_string = lang_word.value()
# -------------------------------------------
# Do some cleaning on the output words
# -------------------------------------------
word_string = clean_lang_token(word_string, lowercase=True)
# For every instance after the first,
# add a space.
out_str = ' {}/{}'
if first:
first = False
out_str = out_str.strip()
output_stream.write(out_str.format(word_string, tag_string))
output_stream.write('\n')
output_stream.flush()
training_sentences += 1
return training_sentences
def gather_gloss_pos_stats(inst, subword_dict, feat_list):
"""
Given an instance, look for the gloss pos tags, and save the statistics
about them, so that we can filter by the number of times each kind was
seen later.
:param inst: Instance to process.
:type inst: RGIgt
:param subword_dict: This dictionary will record the number of times each (word, TAG)
pair has been seen.
:type subword_dict: SubwordDict
:param gram_tag_dict: This dictionary will record the number of times individual grams are seen.
:type gram_tag_dict: TwoLevelCountDict
"""
# Grab the gloss POS tier...
gpos_tier = gloss_tag_tier(inst)
lpos_tier = lang_tag_tier(inst)
gw_tier = gloss(inst)
if CLASS_FEATS_ALN in feat_list:
heur_align_inst(inst)
get_trans_glosses_alignment(inst, aln_method=INTENT_ALN_HEUR)
# If there are POS tags on the language line but not the gloss line...
if gpos_tier is None and lpos_tier is not None:
add_gloss_lang_alignments(inst)
project_lang_to_gloss(inst)
gpos_tier = gloss_tag_tier(inst)
# If this tier exists, then let's process it.
if gpos_tier is not None:
# Iterate over each gloss POS tag...
for i, gw in enumerate(gw_tier):
tag = xigt_find(inst, alignment=gw.id)
if tag is None:
continue
prev_word = gw_tier[i-1].value().lower() if i > 0 else None
next_word = gw_tier[i+1].value().lower() if i < len(gw_tier)-1 else None
if CLASS_FEATS_ALN in feat_list:
subword_dict.add_word_tag(gw.value().lower(), tag.value(), prev_word, next_word)
def evaluate_lang_pos(lang, xc, e_tagger, pos_pla, gold_tagmap=None, outstream=sys.stdout):
"""
:type pos_pla: POSEvalDict
"""
matches = 0
compares = 0
# Iterate through each instance in the corpus.
for inst in xc:
gold_tag_tier = lang_tag_tier(inst, INTENT_POS_MANUAL)
# If there are no gold tags for this instance, skip it.
if gold_tag_tier is None:
continue
# Create the eval tag tier and retrieve it
tag_lang_pos(inst, e_tagger)
eval_tag_tier = lang_tag_tier(inst, INTENT_POS_TAGGER)
# For each gold tag...
for gold_tag in gold_tag_tier:
# Find it's matching tag on the eval side, and compare.
eval_tag = xigt_find(eval_tag_tier, alignment=gold_tag.alignment)
gold_tag_v = gold_tag.value()
if gold_tag_v is not None:
if gold_tagmap:
try:
gold_tag_v = gold_tagmap.get(gold_tag_v)
except TagMapException:
pass
if gold_tag_v != 'JUNK':
if gold_tag_v == eval_tag.value():
matches += 1
compares += 1
pos_pla.add(gold_tag.value(), eval_tag.value())
return matches, compares
def evaluate_pos_projections_on_file(lang, xc, plma, pos_proj_matrix, tagger, gold_tagmap=None, trans_tagmap=None, outstream=sys.stdout):
"""
:type plma: PerLangMethodAccuracies
:type pos_proj_matrix: POSMatrix
"""
new_xc = XigtCorpus(xc.id)
for inst in xc:
gtt = gloss_tag_tier(inst, INTENT_POS_MANUAL)
ttt = trans_tag_tier(inst, INTENT_POS_MANUAL)
m_aln = get_trans_gloss_alignment(inst, INTENT_ALN_MANUAL)
# Only continue if we have manual gloss tags, trans tags, and manual alignment.
if gtt is None or m_aln is None or ttt is None:
continue
# Get the heuristic alignment...
h_aln = heur_align_inst(inst)
# And tag the translation line.
tag_trans_pos(inst, tagger=tagger)
# Now, iterate through each alignment method and set of tags.
for aln_method in [INTENT_ALN_MANUAL, INTENT_ALN_HEUR]:
for trans_tag_method in [INTENT_POS_MANUAL, INTENT_POS_TAGGER]:
project_trans_pos_to_gloss(inst, aln_method=aln_method, trans_tag_method=trans_tag_method)
proj_gtt = gloss_tag_tier(inst, tag_method=INTENT_POS_PROJ)
# Go through each word in the gloss line and, if it has a gold
# tag, was it correct?
matches = 0
compares = 0
for gw in gloss(inst):
gold_tag = xigt_find(gtt, alignment=gw.id)
proj_tag = xigt_find(proj_gtt, alignment=gw.id)
if gold_tag is not None:
gold_tag_v = gold_tag.value()
# Remap the tags if asked...
if gold_tagmap is not None:
try:
gold_tag_v = gold_tagmap.get(gold_tag_v)
except TagMapException:
pass
if proj_tag is None:
proj_str = '**UNK'
else:
proj_str = proj_tag.value()
if trans_tagmap is not None:
# Try to remap the tag, but keep it if it can't be remapped.
try:
proj_str = trans_tagmap.get(proj_str)
except TagMapException:
pass
pos_proj_matrix.add(gold_tag_v, proj_str)
if proj_tag is not None and proj_str == gold_tag_v:
matches += 1
compares += 1
plma.add(lang, '{}:{}'.format(aln_method, trans_tag_method), matches, compares)
outstream.write('{}\n'.format(plma))
return new_xc