Added some runnable applications of this model
This commit is contained in:
parent
4175c45ceb
commit
048825648a
11
.idea/accetuation.iml
generated
11
.idea/accetuation.iml
generated
@ -1,11 +0,0 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<module type="PYTHON_MODULE" version="4">
|
||||
<component name="NewModuleRootManager">
|
||||
<content url="file://$MODULE_DIR$" />
|
||||
<orderEntry type="jdk" jdkName="Python 3.5.2 (~/miniconda3/bin/python)" jdkType="Python SDK" />
|
||||
<orderEntry type="sourceFolder" forTests="false" />
|
||||
</component>
|
||||
<component name="TestRunnerService">
|
||||
<option name="PROJECT_TEST_RUNNER" value="Unittests" />
|
||||
</component>
|
||||
</module>
|
||||
9
.idea/dictionaries/luka.xml
generated
9
.idea/dictionaries/luka.xml
generated
@ -1,9 +0,0 @@
|
||||
<component name="ProjectDictionaryState">
|
||||
<dictionary name="luka">
|
||||
<words>
|
||||
<w>accentuations</w>
|
||||
<w>nonresonant</w>
|
||||
<w>overfitting</w>
|
||||
</words>
|
||||
</dictionary>
|
||||
</component>
|
||||
6
.idea/encodings.xml
generated
6
.idea/encodings.xml
generated
@ -1,6 +0,0 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<project version="4">
|
||||
<component name="Encoding">
|
||||
<file url="PROJECT" charset="UTF-8" />
|
||||
</component>
|
||||
</project>
|
||||
22
.idea/misc.xml
generated
22
.idea/misc.xml
generated
@ -1,22 +0,0 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<project version="4">
|
||||
<component name="ProjectRootManager" version="2" project-jdk-name="Python 3.5.2 (~/miniconda3/bin/python)" project-jdk-type="Python SDK" />
|
||||
<component name="SvnConfiguration">
|
||||
<configuration>$USER_HOME$/.subversion</configuration>
|
||||
</component>
|
||||
<component name="masterDetails">
|
||||
<states>
|
||||
<state key="ScopeChooserConfigurable.UI">
|
||||
<settings>
|
||||
<splitter-proportions>
|
||||
<option name="proportions">
|
||||
<list>
|
||||
<option value="0.2" />
|
||||
</list>
|
||||
</option>
|
||||
</splitter-proportions>
|
||||
</settings>
|
||||
</state>
|
||||
</states>
|
||||
</component>
|
||||
</project>
|
||||
8
.idea/modules.xml
generated
8
.idea/modules.xml
generated
@ -1,8 +0,0 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<project version="4">
|
||||
<component name="ProjectModuleManager">
|
||||
<modules>
|
||||
<module fileurl="file://$PROJECT_DIR$/.idea/accetuation.iml" filepath="$PROJECT_DIR$/.idea/accetuation.iml" />
|
||||
</modules>
|
||||
</component>
|
||||
</project>
|
||||
6
.idea/vcs.xml
generated
6
.idea/vcs.xml
generated
@ -1,6 +0,0 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<project version="4">
|
||||
<component name="VcsDirectoryMappings">
|
||||
<mapping directory="$PROJECT_DIR$" vcs="Git" />
|
||||
</component>
|
||||
</project>
|
||||
1326
.idea/workspace.xml
generated
1326
.idea/workspace.xml
generated
File diff suppressed because it is too large
Load Diff
71
accentuate.py
Normal file
71
accentuate.py
Normal file
@ -0,0 +1,71 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import pickle
|
||||
import numpy as np
|
||||
from keras.models import load_model
|
||||
import sys
|
||||
|
||||
from prepare_data import *
|
||||
|
||||
# obtain data from parameters
|
||||
if len(sys.argv) < 3:
|
||||
print('Please provide arguments for this script to work. First argument should be location of file with unaccented words and morphological data '
|
||||
'and second the name of file where you would like to save file to. Example: python accentuate.py \'test_data/unaccented_dictionary\' '
|
||||
'\'test_data/accented_data\'')
|
||||
raise Exception
|
||||
read_location = sys.argv[1]
|
||||
write_location = sys.argv[2]
|
||||
|
||||
# get environment variables necessary for calculations
|
||||
pickle_input = open('preprocessed_data/environment.pkl', 'rb')
|
||||
environment = pickle.load(pickle_input)
|
||||
dictionary = environment['dictionary']
|
||||
max_word = environment['max_word']
|
||||
max_num_vowels = environment['max_num_vowels']
|
||||
vowels = environment['vowels']
|
||||
accented_vowels = environment['accented_vowels']
|
||||
feature_dictionary = environment['feature_dictionary']
|
||||
syllable_dictionary = environment['syllable_dictionary']
|
||||
|
||||
# load models
|
||||
data = Data('l', shuffle_all_inputs=False)
|
||||
letter_location_model, syllable_location_model, syllabled_letters_location_model = data.load_location_models(
|
||||
'cnn/word_accetuation/cnn_dictionary/v5_3/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllables/v3_3/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllabled_letters/v3_3/20_final_epoch.h5')
|
||||
|
||||
letter_location_co_model, syllable_location_co_model, syllabled_letters_location_co_model = data.load_location_models(
|
||||
'cnn/word_accetuation/cnn_dictionary/v5_2/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllables/v3_2/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllabled_letters/v3_2/20_final_epoch.h5')
|
||||
|
||||
letter_type_model, syllable_type_model, syllabled_letter_type_model = data.load_type_models(
|
||||
'cnn/accent_classification/letters/v3_1/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllables/v2_1/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllabled_letters/v2_1/20_final_epoch.h5')
|
||||
|
||||
letter_type_co_model, syllable_type_co_model, syllabled_letter_type_co_model = data.load_type_models(
|
||||
'cnn/accent_classification/letters/v3_0/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllables/v2_0/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllabled_letters/v2_0/20_final_epoch.h5')
|
||||
|
||||
# read from data
|
||||
content = data._read_content(read_location)
|
||||
|
||||
# format data for accentuate_word function it has to be like [['besedišči', '', 'Ncnpi', 'besedišči'], ]
|
||||
content = [[el[0], '', el[1][:-1], el[0]] for el in content[:-1]]
|
||||
|
||||
# use environment variables and models to accentuate words
|
||||
data = Data('l', shuffle_all_inputs=False)
|
||||
location_accented_words, accented_words = data.accentuate_word(content, letter_location_model, syllable_location_model, syllabled_letters_location_model,
|
||||
letter_location_co_model, syllable_location_co_model, syllabled_letters_location_co_model,
|
||||
letter_type_model, syllable_type_model, syllabled_letter_type_model,
|
||||
letter_type_co_model, syllable_type_co_model, syllabled_letter_type_co_model,
|
||||
dictionary, max_word, max_num_vowels, vowels, accented_vowels, feature_dictionary, syllable_dictionary)
|
||||
|
||||
# save accentuated words
|
||||
with open(write_location, 'w') as f:
|
||||
for i in range(len(location_accented_words)):
|
||||
f.write(location_accented_words[i] + ' ' + accented_words[i] + '\n')
|
||||
f.write('\n')
|
||||
79
accentuate_connected_text.py
Normal file
79
accentuate_connected_text.py
Normal file
@ -0,0 +1,79 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import sys
|
||||
|
||||
sys.path.insert(0, '../../../')
|
||||
from prepare_data import *
|
||||
|
||||
import pickle
|
||||
|
||||
# from keras import backend as Input
|
||||
np.random.seed(7)
|
||||
|
||||
# obtain data from parameters
|
||||
if len(sys.argv) < 3:
|
||||
print('Please provide arguments for this script to work. First argument should be location of file with unaccented words and morphological data, '
|
||||
'second the name of file where you would like to save results to and third location of ReLDI tagger. Example: python accentuate.py '
|
||||
'\'test_data/original_connected_text\' \'test_data/accented_connected_text\' \'../reldi_tagger\'')
|
||||
raise Exception
|
||||
read_location = sys.argv[1]
|
||||
write_location = sys.argv[2]
|
||||
reldi_location = sys.argv[3]
|
||||
|
||||
# get environment variables necessary for calculations
|
||||
pickle_input = open('preprocessed_data/environment.pkl', 'rb')
|
||||
environment = pickle.load(pickle_input)
|
||||
dictionary = environment['dictionary']
|
||||
max_word = environment['max_word']
|
||||
max_num_vowels = environment['max_num_vowels']
|
||||
vowels = environment['vowels']
|
||||
accented_vowels = environment['accented_vowels']
|
||||
feature_dictionary = environment['feature_dictionary']
|
||||
syllable_dictionary = environment['syllable_dictionary']
|
||||
|
||||
# get models
|
||||
data = Data('l', shuffle_all_inputs=False)
|
||||
letter_location_model, syllable_location_model, syllabled_letters_location_model = data.load_location_models(
|
||||
'cnn/word_accetuation/cnn_dictionary/v5_3/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllables/v3_3/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllabled_letters/v3_3/20_final_epoch.h5')
|
||||
|
||||
letter_location_co_model, syllable_location_co_model, syllabled_letters_location_co_model = data.load_location_models(
|
||||
'cnn/word_accetuation/cnn_dictionary/v5_2/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllables/v3_2/20_final_epoch.h5',
|
||||
'cnn/word_accetuation/syllabled_letters/v3_2/20_final_epoch.h5')
|
||||
|
||||
letter_type_model, syllable_type_model, syllabled_letter_type_model = data.load_type_models(
|
||||
'cnn/accent_classification/letters/v3_1/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllables/v2_1/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllabled_letters/v2_1/20_final_epoch.h5')
|
||||
|
||||
letter_type_co_model, syllable_type_co_model, syllabled_letter_type_co_model = data.load_type_models(
|
||||
'cnn/accent_classification/letters/v3_0/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllables/v2_0/20_final_epoch.h5',
|
||||
'cnn/accent_classification/syllabled_letters/v2_0/20_final_epoch.h5')
|
||||
|
||||
# get word tags
|
||||
tagged_words, original_text = data.tag_words(reldi_location, read_location)
|
||||
|
||||
# find accentuation locations
|
||||
predictions = data.get_ensemble_location_predictions(tagged_words, letter_location_model, syllable_location_model, syllabled_letters_location_model,
|
||||
letter_location_co_model, syllable_location_co_model, syllabled_letters_location_co_model,
|
||||
dictionary, max_word, max_num_vowels, vowels, accented_vowels, feature_dictionary,
|
||||
syllable_dictionary)
|
||||
|
||||
location_accented_text = data.create_connected_text_locations(tagged_words, original_text, predictions, vowels)
|
||||
|
||||
# accentuate text
|
||||
location_y = np.around(predictions)
|
||||
type_predictions = data.get_ensemble_type_predictions(tagged_words, location_y, letter_type_model, syllable_type_model, syllabled_letter_type_model,
|
||||
letter_type_co_model, syllable_type_co_model, syllabled_letter_type_co_model,
|
||||
dictionary, max_word, max_num_vowels, vowels, accented_vowels, feature_dictionary,
|
||||
syllable_dictionary)
|
||||
|
||||
accented_text = data.create_connected_text_accented(tagged_words, original_text, type_predictions, location_y, vowels, accented_vowels)
|
||||
|
||||
# save accentuated text
|
||||
with open(write_location, 'w') as f:
|
||||
f.write(accented_text)
|
||||
74
learn_location_weights.py
Normal file
74
learn_location_weights.py
Normal file
@ -0,0 +1,74 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
from __future__ import unicode_literals
|
||||
# text in Western (Windows 1252)
|
||||
|
||||
import pickle
|
||||
import numpy as np
|
||||
np.random.seed(7)
|
||||
|
||||
import sys
|
||||
from prepare_data import *
|
||||
|
||||
# preprocess data
|
||||
# data = Data('l', allow_shuffle_vector_generation=True, save_generated_data=False, shuffle_all_inputs=True)
|
||||
data = Data('l', save_generated_data=False, shuffle_all_inputs=True)
|
||||
data.generate_data('../../internal_representations/inputs/letters_word_accentuation_train',
|
||||
'../../internal_representations/inputs/letters_word_accentuation_test',
|
||||
'../../internal_representations/inputs/letters_word_accentuation_validate',
|
||||
content_location='../accetuation/data/',
|
||||
content_name='SlovarIJS_BESEDE_utf8.lex',
|
||||
inputs_location='../accetuation/cnn/internal_representations/inputs/',
|
||||
content_shuffle_vector='content_shuffle_vector',
|
||||
shuffle_vector='shuffle_vector')
|
||||
|
||||
# combine all data (if it is unwanted comment code below)
|
||||
data.x_train = np.concatenate((data.x_train, data.x_test, data.x_validate), axis=0)
|
||||
data.x_other_features_train = np.concatenate((data.x_other_features_train, data.x_other_features_test, data.x_other_features_validate), axis=0)
|
||||
data.y_train = np.concatenate((data.y_train, data.y_test, data.y_validate), axis=0)
|
||||
|
||||
# build neural network architecture
|
||||
nn_output_dim = 10
|
||||
batch_size = 16
|
||||
actual_epoch = 20
|
||||
num_fake_epoch = 20
|
||||
|
||||
conv_input_shape=(23, 36)
|
||||
othr_input = (140, )
|
||||
|
||||
conv_input = Input(shape=conv_input_shape, name='conv_input')
|
||||
x_conv = Conv1D(115, (3), padding='same', activation='relu')(conv_input)
|
||||
x_conv = Conv1D(46, (3), padding='same', activation='relu')(x_conv)
|
||||
x_conv = MaxPooling1D(pool_size=2)(x_conv)
|
||||
x_conv = Flatten()(x_conv)
|
||||
|
||||
othr_input = Input(shape=othr_input, name='othr_input')
|
||||
|
||||
x = concatenate([x_conv, othr_input])
|
||||
x = Dense(256, activation='relu')(x)
|
||||
x = Dropout(0.3)(x)
|
||||
x = Dense(256, activation='relu')(x)
|
||||
x = Dropout(0.3)(x)
|
||||
x = Dense(256, activation='relu')(x)
|
||||
x = Dropout(0.3)(x)
|
||||
x = Dense(nn_output_dim, activation='sigmoid')(x)
|
||||
|
||||
model = Model(inputs=[conv_input, othr_input], outputs=x)
|
||||
opt = optimizers.Adam(lr=1E-3, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
|
||||
model.compile(loss='mean_squared_error', optimizer=opt, metrics=[actual_accuracy,])
|
||||
# model.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
|
||||
|
||||
|
||||
# start learning
|
||||
history = model.fit_generator(data.generator('train', batch_size, content_name='SlovarIJS_BESEDE_utf8.lex', content_location='../accetuation/data/'),
|
||||
data.x_train.shape[0]/(batch_size * num_fake_epoch),
|
||||
epochs=actual_epoch*num_fake_epoch,
|
||||
validation_data=data.generator('test', batch_size),
|
||||
validation_steps=data.x_test.shape[0]/(batch_size * num_fake_epoch))
|
||||
|
||||
|
||||
# save generated data
|
||||
name = 'test_data/20_epoch'
|
||||
model.save(name + '.h5')
|
||||
output = open(name + '_history.pkl', 'wb')
|
||||
pickle.dump(history.history, output)
|
||||
output.close()
|
||||
@ -7,6 +7,7 @@ import h5py
|
||||
import math
|
||||
import keras.backend as K
|
||||
import os.path
|
||||
from os import remove
|
||||
import codecs
|
||||
|
||||
from copy import copy
|
||||
@ -666,7 +667,7 @@ class Data:
|
||||
loc += batch_size
|
||||
|
||||
# generator for inputs for tracking of data fitting
|
||||
def _syllable_generator(self, orig_x, orig_x_additional, orig_y, batch_size, translator, accented_vowels, oversampling):
|
||||
def _syllable_generator(self, orig_x, orig_x_additional, orig_y, batch_size, translator, accented_vowels, oversampling=np.ones(13)):
|
||||
size = orig_x.shape[0]
|
||||
while 1:
|
||||
loc = 0
|
||||
@ -1655,6 +1656,95 @@ class Data:
|
||||
|
||||
return location_accented_words, accented_words
|
||||
|
||||
def tag_words(self, reldi_location, original_location):
|
||||
# generates text with every word in new line
|
||||
with open(original_location) as f:
|
||||
original_text = f.readlines()
|
||||
original_text = ''.join(original_text)
|
||||
# print(original_text)
|
||||
text_with_whitespaces = original_text.replace(',', ' ,').replace('.', ' .').replace('\n', ' ').replace("\"", " \" ").replace(":",
|
||||
" :").replace(
|
||||
"ć", "č").replace('–', '-')
|
||||
# print('-------------------------------------------------')
|
||||
text_with_whitespaces = '\n'.join(text_with_whitespaces.split())
|
||||
text_with_whitespaces += '\n\n'
|
||||
# print(text_with_whitespaces)
|
||||
with open('.words_with_whitespaces', "w") as text_file:
|
||||
text_file.write(text_with_whitespaces)
|
||||
|
||||
# generates text with PoS tags
|
||||
import subprocess
|
||||
|
||||
myinput = open('.words_with_whitespaces', 'r')
|
||||
myoutput = open('.word_tags', 'w')
|
||||
# print(myinput.readlines())
|
||||
python3_command = reldi_location + "/tagger.py sl" # launch your python2 script using bash
|
||||
|
||||
process = subprocess.run(python3_command.split(), stdin=myinput, stdout=myoutput)
|
||||
|
||||
# generates interesting words
|
||||
pointless_words = ['.', ',', '\"', ':', '-']
|
||||
with open('.word_tags', "r") as text_file:
|
||||
tagged_input_words = []
|
||||
for x in text_file.readlines()[:-1]:
|
||||
splited_line = x[:-1].split('\t')
|
||||
if splited_line[0] not in pointless_words and not any(char.isdigit() for char in splited_line[0]):
|
||||
tagged_input_words.append([splited_line[0].lower(), '', splited_line[1], splited_line[0].lower()])
|
||||
|
||||
remove(".words_with_whitespaces")
|
||||
remove(".word_tags")
|
||||
return tagged_input_words, original_text
|
||||
|
||||
def create_connected_text_locations(self, tagged_input_words, original_text, predictions, vowels):
|
||||
if 'A' not in vowels:
|
||||
vowels.extend(['A', 'E', 'I', 'O', 'U'])
|
||||
accented_words = [self.assign_location_stress(tagged_input_words[i][0][::-1], self.decode_y(predictions[i]), vowels)[::-1] for i in
|
||||
range(len(tagged_input_words))]
|
||||
|
||||
# print(accented_words[:20])
|
||||
# print(tagged_input_words[:20])
|
||||
|
||||
words_and_accetuation_loc = [[tagged_input_words[i][0], self.decode_y(predictions[i])] for i in range(len(tagged_input_words))]
|
||||
|
||||
original_text_list = list(original_text)
|
||||
original_text_lowercase = original_text.lower()
|
||||
end_pos = 0
|
||||
for word in words_and_accetuation_loc:
|
||||
posit = original_text_lowercase.find(word[0], end_pos)
|
||||
if posit != -1:
|
||||
start_pos = posit
|
||||
end_pos = start_pos + len(word[0])
|
||||
|
||||
original_text_list[start_pos:end_pos] = list(
|
||||
self.assign_location_stress(''.join(original_text_list[start_pos:end_pos][::-1]), word[1], vowels)[::-1])
|
||||
|
||||
return ''.join(original_text_list)
|
||||
|
||||
def create_connected_text_accented(self, tagged_input_words, original_text, type_predictions, location_y, vowels, accented_vowels):
|
||||
|
||||
input_words = [el[0] for el in tagged_input_words]
|
||||
words = self.assign_stress_types(type_predictions, input_words, location_y, vowels, accented_vowels)
|
||||
|
||||
# print(original_text)
|
||||
|
||||
original_text_list = list(original_text)
|
||||
original_text_lowercase = original_text.lower()
|
||||
end_pos = 0
|
||||
for i in range(len(words)):
|
||||
posit = original_text_lowercase.find(input_words[i], end_pos)
|
||||
if posit != -1:
|
||||
start_pos = posit
|
||||
end_pos = start_pos + len(words[i])
|
||||
|
||||
orig_word = original_text_list[start_pos:end_pos]
|
||||
new_word = list(words[i])
|
||||
for j in range(len(orig_word)):
|
||||
if orig_word[j].isupper():
|
||||
new_word[j] = new_word[j].upper()
|
||||
|
||||
original_text_list[start_pos:end_pos] = new_word
|
||||
|
||||
return ''.join(original_text_list)
|
||||
# def count_vowels(content, vowels):
|
||||
# num_all_vowels = 0
|
||||
# for el in content:
|
||||
|
||||
BIN
preprocessed_data/environment.pkl
Normal file
BIN
preprocessed_data/environment.pkl
Normal file
Binary file not shown.
1
test_data/accented_connected_text
Normal file
1
test_data/accented_connected_text
Normal file
@ -0,0 +1 @@
|
||||
Izbrúhi na sóncu só žé vëčkrat pokazáli zóbe nášim satelítom, poslédično nášim mobílnim telefónom, navigáciji, celo eléktričnemu omréžju. Á vesóljskega vreména šë në morémo napovédati – kakó bî ga láhko, se tá téden na Blédu pogovárja okóli 70 znánstvenikov Evrópske vesóljske agéncije, ki jé sebój pripeljála svôjo näjvéčjo ikóno, británca Mátta Taylorja.
|
||||
6
test_data/accented_data
Normal file
6
test_data/accented_data
Normal file
@ -0,0 +1,6 @@
|
||||
absolutístični absolutístični
|
||||
spoštljívejše spoštljívejše
|
||||
tresóče tresóče
|
||||
razneséna raznesěna
|
||||
žvížgih žvížgih
|
||||
|
||||
1
test_data/original_connected_text
Normal file
1
test_data/original_connected_text
Normal file
@ -0,0 +1 @@
|
||||
Izbruhi na soncu so že večkrat pokazali zobe našim satelitom, posledično našim mobilnim telefonom, navigaciji, celo električnemu omrežju. A vesoljskega vremena še ne moremo napovedati – kako bi ga lahko, se ta teden na Bledu pogovarja okoli 70 znanstvenikov Evropske vesoljske agencije, ki je seboj pripeljala svojo največjo ikono, britanca Matta Taylorja.
|
||||
6
test_data/unaccented_dictionary
Normal file
6
test_data/unaccented_dictionary
Normal file
@ -0,0 +1,6 @@
|
||||
absolutistični Afpmsay-n
|
||||
spoštljivejše Afcfsg
|
||||
tresoče Afpfsg
|
||||
raznesena Vmp--sfp
|
||||
žvižgih Ncmdl
|
||||
|
||||
Loading…
Reference in New Issue
Block a user