def test_phonetic_edit_distance_alignment(self):
"""Test abydos.distance.PhoneticEditDistance.alignment."""
# Base cases
self.assertEqual(self.ped.alignment('', ''), (0.0, '', ''))
self.assertEqual(self.ped.alignment('a', ''), (1.0, 'a', '-'))
self.assertEqual(self.ped.alignment('', 'a'), (1.0, '-', 'a'))
self.assertEqual(self.ped.alignment('abc', ''), (3.0, 'abc', '---'))
self.assertEqual(self.ped.alignment('', 'abc'), (3.0, '---', 'abc'))
self.assertEqual(self.ped.alignment('abc', 'abc'), (0.0, 'abc', 'abc'))
self.assertEqual(
self.ped.alignment('abcd', 'efgh'),
(0.4193548387096774, 'abcd', 'efgh'),
)
self.assertEqual(
self.ped.alignment('Nigel', 'Niall'),
(0.8870967741935485, 'Nigel', 'Niall'),
)
self.assertEqual(
self.ped.alignment('Niall', 'Nigel'),
(0.8870967741935485, 'Niall', 'Nigel'),
)
self.assertEqual(
self.ped.alignment('Colin', 'Coiln'),
(0.870967741935484, 'Colin', 'Coiln'),
)
self.assertEqual(
self.ped.alignment('Coiln', 'Colin'),
(0.870967741935484, 'Coiln', 'Colin'),
)
self.assertEqual(
PhoneticEditDistance(mode='osa').alignment('Niel', 'Neil'),
(0.06451612903225801, 'Niel', 'Neil'),
)
def test_phonetic_edit_distance_dist_abs(self):
"""Test abydos.distance.PhoneticEditDistance.dist_abs."""
# Base cases
self.assertEqual(self.ped.dist_abs('', ''), 0)
self.assertEqual(self.ped.dist_abs('a', ''), 1)
self.assertEqual(self.ped.dist_abs('', 'a'), 1)
self.assertEqual(self.ped.dist_abs('abc', ''), 3)
self.assertEqual(self.ped.dist_abs('', 'abc'), 3)
self.assertEqual(self.ped.dist_abs('abc', 'abc'), 0)
self.assertEqual(self.ped.dist_abs('abcd', 'efgh'), 0.4193548387096774)
self.assertAlmostEqual(self.ped.dist_abs('Nigel', 'Niall'),
0.8870967741935485)
self.assertAlmostEqual(self.ped.dist_abs('Niall', 'Nigel'),
0.8870967741935485)
self.assertAlmostEqual(self.ped.dist_abs('Colin', 'Coiln'),
0.870967741935484)
self.assertAlmostEqual(self.ped.dist_abs('Coiln', 'Colin'),
0.870967741935484)
self.assertAlmostEqual(self.ped.dist_abs('ATCAACGAGT', 'AACGATTAG'),
2.370967741935484)
self.assertEqual(
PhoneticEditDistance(weights={
'syllabic': 1.0
}).dist_abs('Nigel', 'Niall'),
0.0,
)
self.assertAlmostEqual(
PhoneticEditDistance(weights=(1, 1, 1)).dist_abs('Nigel', 'Niall'),
0.33333333333333326,
)
self.assertAlmostEqual(
PhoneticEditDistance(mode='osa').dist_abs('Niel', 'Neil'),
0.06451612903225801,
)
def __init__(self, model='latin'):
self.model = model
self.impH = input_helpers.InputHelper()
self.ST = syllable_tokenizer.SyllableTokenizer()
# Phonetic Encoder
self.pe = Ainsworth()
# Soundex Firstname Algorithm
self.pshp_soundex_first = PSHPSoundexFirst()
# String Distance algorithms
self.algos = [
IterativeSubString(),
BISIM(),
DiscountedLevenshtein(),
Prefix(),
LCSstr(),
MLIPNS(),
Strcmp95(),
MRA(),
Editex(),
SAPS(),
FlexMetric(),
JaroWinkler(mode='Jaro'),
HigueraMico(),
Sift4(),
Eudex(),
ALINE(),
Covington(),
PhoneticEditDistance()
]
self.algo_names = [
'iterativesubstring', 'bisim', 'discountedlevenshtein', 'prefix',
'lcsstr', 'mlipns', 'strcmp95', 'mra', 'editex', 'saps',
'flexmetric', 'jaro', 'higueramico', 'sift4', 'eudex', 'aline',
'covington', 'phoneticeditdistance'
]
# extract model tarball into directory if doesnt exist
model_dir = os.path.join(os.path.dirname(__file__), "models",
self.model)
if not os.path.exists(model_dir):
os.makedirs(model_dir)
tar = tarfile.open(
os.path.join(os.path.dirname(__file__), "models",
self.model + ".tar.gz"), "r:gz")
tar.extractall(model_dir)
tar.close()
# String Distance Pipeline (Level 0/Base Model)
self.baseModel = joblib.load(os.path.join(model_dir, 'base.pkl'))
# Character Embedding Network (Level 0/Base Model)
self.vocab = preprocess.VocabularyProcessor(
max_document_length=15,
min_frequency=0).restore(os.path.join(model_dir, 'vocab'))
siamese_model = os.path.join(model_dir, 'siamese')
graph = tf.Graph()
with graph.as_default() as graph:
self.sess = tf.Session()
with self.sess.as_default():
# Load the saved meta graph and restore variables
saver = tf.train.import_meta_graph(
'{}.meta'.format(siamese_model))
self.sess.run(tf.global_variables_initializer())
saver.restore(self.sess, siamese_model)
# Get the placeholders from the graph by name
self.input_x1 = graph.get_operation_by_name('input_x1').outputs[0]
self.input_x2 = graph.get_operation_by_name('input_x2').outputs[0]
self.dropout_keep_prob = graph.get_operation_by_name(
'dropout_keep_prob').outputs[0]
self.prediction = graph.get_operation_by_name(
'output/distance').outputs[0]
self.sim = graph.get_operation_by_name(
'accuracy/temp_sim').outputs[0]
# Logreg (Level 1/Meta Model)
self.metaModel = joblib.load(os.path.join(model_dir, 'meta.pkl'))
# seen names (mapping dict from raw name to processed name)
self.seen_names = {}
# seen pairs (mapping dict from name pair tuple to similarity)
self.seen_pairs = {}
bisim = BISIM()
dlev = DiscountedLevenshtein()
prefix = Prefix()
lcs = LCSstr()
mlipns = MLIPNS()
strcmp95 = Strcmp95()
mra = MRA()
editex = Editex()
saps = SAPS()
flexmetric = FlexMetric()
jaro = JaroWinkler(mode='Jaro')
higuera_mico = HigueraMico()
sift4 = Sift4()
eudex = Eudex()
aline = ALINE()
phonetic_edit = PhoneticEditDistance()
algos = [iss, bisim, dlev, prefix, lcs, mlipns, strcmp95, mra, editex, saps, flexmetric, jaro, higuera_mico, sift4, eudex,
aline, phonetic_edit]
algo_names = ['iterativesubstring', 'bisim', 'discountedlevenshtein', 'prefix', 'lcsstr', 'mlipns', 'strcmp95', 'mra',
'editex', 'saps', 'flexmetric', 'jaro', 'higueramico', 'sift4', 'eudex', 'aline',
'phoneticeditdistance']
def sum_ipa(name_a, name_b):
feat1 = ipa_to_features(pe.encode(name_a))
feat2 = ipa_to_features(pe.encode(name_b))
if len(feat1) <= 1:
score = sum(cmp_features(f1, f2) for f1, f2 in zip(feat1, feat2))/1
else:
score = sum(cmp_features(f1, f2) for f1, f2 in zip(feat1, feat2))/len(feat1)
return score
def __init__(self, model='latin', prefilter=True, allow_alt_surname=True, allow_initials=True,
allow_missing_components=True):
# user-provided parameters
self.model = model
self.allow_alt_surname = allow_alt_surname
self.allow_initials = allow_initials
self.allow_missing_components = allow_missing_components
self.prefilter = prefilter
if self.prefilter:
self.refined_soundex = {
'b': 1, 'p': 1,
'f': 2, 'v': 2,
'c': 3, 'k': 3, 's': 3,
'g': 4, 'j': 4,
'q': 5, 'x': 5, 'z': 5,
'd': 6, 't': 6,
'l': 7,
'm': 8, 'n': 8,
'r': 9
}
# verify user-supplied class arguments
model_dir = self.validate_parameters()
self.impH = input_helpers.InputHelper()
# Phonetic Encoder
self.pe = Ainsworth()
# Soundex Firstname Algorithm
self.pshp_soundex_first = PSHPSoundexFirst()
# Soundex Lastname Algorithm
self.pshp_soundex_last = PSHPSoundexLast()
# String Distance algorithms
self.algos = [IterativeSubString(), BISIM(), DiscountedLevenshtein(), Prefix(), LCSstr(), MLIPNS(),
Strcmp95(), MRA(), Editex(), SAPS(), FlexMetric(), JaroWinkler(mode='Jaro'), HigueraMico(),
Sift4(), Eudex(), ALINE(), CovingtonGuard(), PhoneticEditDistance()]
self.algo_names = ['iterativesubstring', 'bisim', 'discountedlevenshtein', 'prefix', 'lcsstr', 'mlipns',
'strcmp95', 'mra', 'editex', 'saps', 'flexmetric', 'jaro', 'higueramico',
'sift4', 'eudex', 'aline', 'covington', 'phoneticeditdistance']
# String Distance Pipeline (Level 0/Base Model)
self.baseModel = joblib.load(os.path.join(model_dir, 'base.pkl'))
# Character Embedding Network (Level 0/Base Model)
self.vocab = preprocess.VocabularyProcessor(max_document_length=15, min_frequency=0).restore(
os.path.join(model_dir, 'vocab'))
siamese_model = os.path.join(model_dir, 'siamese')
# start tensorflow session
graph = tf.Graph()
with graph.as_default() as graph:
self.sess = tf.Session() if tf.__version__[0] == '1' else tf.compat.v1.Session()
with self.sess.as_default():
# Load the saved meta graph and restore variables
if tf.__version__[0] == '1':
saver = tf.train.import_meta_graph('{}.meta'.format(siamese_model))
self.sess.run(tf.global_variables_initializer())
else:
saver = tf.compat.v1.train.import_meta_graph('{}.meta'.format(siamese_model))
self.sess.run(tf.compat.v1.global_variables_initializer())
saver.restore(self.sess, siamese_model)
# Get the placeholders from the graph by name
self.input_x1 = graph.get_operation_by_name('input_x1').outputs[0]
self.input_x2 = graph.get_operation_by_name('input_x2').outputs[0]
self.dropout_keep_prob = graph.get_operation_by_name('dropout_keep_prob').outputs[0]
self.prediction = graph.get_operation_by_name('output/distance').outputs[0]
self.sim = graph.get_operation_by_name('accuracy/temp_sim').outputs[0]
# Logreg (Level 1/Meta Model)
self.metaModel = joblib.load(os.path.join(model_dir, 'meta.pkl'))
# seen names (mapping dict from raw name to processed name)
self.seen_names = {}
# seen pairs (mapping dict from name pair tuple to similarity)
self.seen_pairs = {}
# user scores (mapping dict from name pair tuple to similarity)
self.user_scores = {}
"""Evaluation functions for sequence models."""
__author__ = "Kyle Gorman"
import logging
import numpy # type: ignore
from typing import Any, Iterator, List, Tuple
from abydos.distance import PhoneticEditDistance
Labels = List[Any]
Metrics = [
PhoneticEditDistance(vowel_ignorance=True),
PhoneticEditDistance(vowel_dominance=True),
PhoneticEditDistance(vowel_ignorance=True, no_features=True),
PhoneticEditDistance()
]
def edit_distance(x: Labels, y: Labels) -> int:
# For a more expressive version of the same, see:
#
# https://gist.github.com/kylebgorman/8034009
idim = len(x) + 1
jdim = len(y) + 1
table = numpy.zeros((idim, jdim), dtype=numpy.uint8)
table[1:, 0] = 1
table[0, 1:] = 1
import argparse
from abydos.distance import Levenshtein, PhoneticEditDistance
from evallib import edit_distance
from tqdm import tqdm
parser = argparse.ArgumentParser()
parser.add_argument('--left', type=str)
parser.add_argument('--right', type=str)
parser.add_argument('--out', type=str)
args = parser.parse_args()
PED = PhoneticEditDistance()
PED_IG = PhoneticEditDistance(vowel_ignorance=True)
PED_DOM = PhoneticEditDistance(vowel_dominance=True)
PED_NOF_IG = PhoneticEditDistance(vowel_ignorance=True, no_features=True)
PED_NOF = PhoneticEditDistance(no_features=True)
LEV = Levenshtein()
with open(args.left) as left, open(args.right) as right, open(args.out,
'w') as ouf:
for left_line, right_line in tqdm(zip(left, right)):
left_grapheme, left_phoneme = left_line.strip().split('\t')
right_grapheme, right_phoneme = right_line.strip().split('\t')
lsplit = left_phoneme.split()
rsplit = right_phoneme.split()
gorman_lev = edit_distance(lsplit, rsplit) / len(lsplit)
left_phon_input = ''.join(lsplit)
class PhoneticEditDistanceTestCases(unittest.TestCase):
"""Test phonetic edit distance functions.
abydos.distance.PhoneticEditDistance
"""
ped = PhoneticEditDistance()
def test_phonetic_edit_distance_dist(self):
"""Test abydos.distance.PhoneticEditDistance.dist."""
# Base cases
self.assertEqual(self.ped.dist('', ''), 0.0)
self.assertEqual(self.ped.dist('a', ''), 1.0)
self.assertEqual(self.ped.dist('', 'a'), 1.0)
self.assertEqual(self.ped.dist('abc', ''), 1.0)
self.assertEqual(self.ped.dist('', 'abc'), 1.0)
self.assertEqual(self.ped.dist('abc', 'abc'), 0.0)
self.assertEqual(self.ped.dist('abcd', 'efgh'), 0.10483870967741934)
self.assertAlmostEqual(self.ped.dist('Nigel', 'Niall'),
0.1774193548387097)
self.assertAlmostEqual(self.ped.dist('Niall', 'Nigel'),
0.1774193548387097)
self.assertAlmostEqual(self.ped.dist('Colin', 'Coiln'),
0.1741935483870968)
self.assertAlmostEqual(self.ped.dist('Coiln', 'Colin'),
0.1741935483870968)
self.assertAlmostEqual(self.ped.dist('ATCAACGAGT', 'AACGATTAG'),
0.2370967741935484)
def test_phonetic_edit_distance_dist_abs(self):
"""Test abydos.distance.PhoneticEditDistance.dist_abs."""
# Base cases
self.assertEqual(self.ped.dist_abs('', ''), 0)
self.assertEqual(self.ped.dist_abs('a', ''), 1)
self.assertEqual(self.ped.dist_abs('', 'a'), 1)
self.assertEqual(self.ped.dist_abs('abc', ''), 3)
self.assertEqual(self.ped.dist_abs('', 'abc'), 3)
self.assertEqual(self.ped.dist_abs('abc', 'abc'), 0)
self.assertEqual(self.ped.dist_abs('abcd', 'efgh'), 0.4193548387096774)
self.assertAlmostEqual(self.ped.dist_abs('Nigel', 'Niall'),
0.8870967741935485)
self.assertAlmostEqual(self.ped.dist_abs('Niall', 'Nigel'),
0.8870967741935485)
self.assertAlmostEqual(self.ped.dist_abs('Colin', 'Coiln'),
0.870967741935484)
self.assertAlmostEqual(self.ped.dist_abs('Coiln', 'Colin'),
0.870967741935484)
self.assertAlmostEqual(self.ped.dist_abs('ATCAACGAGT', 'AACGATTAG'),
2.370967741935484)
self.assertEqual(
PhoneticEditDistance(weights={
'syllabic': 1.0
}).dist_abs('Nigel', 'Niall'),
0.0,
)
self.assertAlmostEqual(
PhoneticEditDistance(weights=(1, 1, 1)).dist_abs('Nigel', 'Niall'),
0.33333333333333326,
)
self.assertAlmostEqual(
PhoneticEditDistance(mode='osa').dist_abs('Niel', 'Neil'),
0.06451612903225801,
)
def test_phonetic_edit_distance_alignment(self):
"""Test abydos.distance.PhoneticEditDistance.alignment."""
# Base cases
self.assertEqual(self.ped.alignment('', ''), (0.0, '', ''))
self.assertEqual(self.ped.alignment('a', ''), (1.0, 'a', '-'))
self.assertEqual(self.ped.alignment('', 'a'), (1.0, '-', 'a'))
self.assertEqual(self.ped.alignment('abc', ''), (3.0, 'abc', '---'))
self.assertEqual(self.ped.alignment('', 'abc'), (3.0, '---', 'abc'))
self.assertEqual(self.ped.alignment('abc', 'abc'), (0.0, 'abc', 'abc'))
self.assertEqual(
self.ped.alignment('abcd', 'efgh'),
(0.4193548387096774, 'abcd', 'efgh'),
)
self.assertEqual(
self.ped.alignment('Nigel', 'Niall'),
(0.8870967741935485, 'Nigel', 'Niall'),
)
self.assertEqual(
self.ped.alignment('Niall', 'Nigel'),
(0.8870967741935485, 'Niall', 'Nigel'),
)
self.assertEqual(
self.ped.alignment('Colin', 'Coiln'),
(0.870967741935484, 'Colin', 'Coiln'),
)
self.assertEqual(
self.ped.alignment('Coiln', 'Colin'),
(0.870967741935484, 'Coiln', 'Colin'),
)
self.assertEqual(
PhoneticEditDistance(mode='osa').alignment('Niel', 'Neil'),
(0.06451612903225801, 'Niel', 'Neil'),
)