def grammemes_xml2csv(self, persistent=True):
env = Environment()
filename_gram = env.filename_grammemes_xml()
dfcols = ['name', 'alias', 'description']
df_xml = pd.DataFrame(columns=dfcols)
try:
tree = ET.ElementTree(file=filename_gram)
except:
env.debug(1, ['Failed to load grammemes from XML:', filename_gram])
else:
env.debug(1, ['Read grammemes:', filename_gram])
for elem in tree.iter('grammeme'):
#print(elem.tag, elem.attrib)
sattr = elem.attrib.get('include')
if sattr == 'on':
sname = sali = sdesc = ''
for child in elem:
if child.tag.lower() == 'name':
sname = child.text.upper()
elif child.tag.lower() == 'alias':
sali = child.text.upper()
elif child.tag.lower() == 'description':
sdesc = child.text.lower()
s = pd.Series(data=[sname, sali, sdesc], index=dfcols)
df_xml = df_xml.append(s, ignore_index=True)
df_xml.index.name = 'idgram'
if persistent:
filename_csv = env.filename_grammemes_csv()
env.debug(1, ['Write grammemes to CSV:', filename_csv])
df_xml.to_csv(filename_csv, encoding='utf-8')
return df_xml
def vocabulary_from_corpus(self, n_min=1, n_max=10, persistent=True):
env = Environment()
df_voc = pd.DataFrame()
#dfgram = self.grammemes()
for i in range(n_min, n_max + 1):
file_csv = env.filename_corpus_csv(i)
try:
dffile = pd.read_csv(file_csv,
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(1, ['Failed to read corpus file:', file_csv])
else:
env.debug(1, ['Read OK:', file_csv])
if not dffile.empty:
df_voc = df_voc.append(dffile)
df_voc = df_voc.drop_duplicates()
df_voc.columns = ['word', 'gram', 'idgram']
df_voc = df_voc.reset_index(drop=True)
df_voc.index.name = 'idcorpus'
if persistent:
file_voc = env.filename_vocabulary_csv()
env.debug(1, ['Write vocabulary to CSV:', file_voc])
df_voc.to_csv(file_voc, encoding='utf-8')
return df_voc
def get_texts_stat(self, mode='train'):
# Готовим данные
env = Environment()
if mode == 'train':
file_res = env.filename_results_csv()
if mode == 'test':
file_res = env.filename_stat_test_csv()
authors = pd.read_csv(env.filename_authors_csv(),
index_col='idauthor',
encoding='utf-8')
data = pd.read_csv(file_res, index_col='idstat', encoding='utf-8')
data.drop(columns=['file', 'idchunk'], inplace=True)
columns = data.columns
group = data.groupby(['idtext', 'idauthor', 'author', 'name'])
group = group.agg({
'sentences_text': ['mean'],
'words_text': ['mean'],
'sentence_mean': ['mean'],
'sentences_chunk': ['mean'],
'words_chunk': ['mean'],
'words_uniq_chunk': ['mean'],
'uniq_per_sent_chunk': ['mean'],
'uniq_per_words_chunk': ['mean'],
'NOUN': ['mean'],
'ADJF': ['mean'],
'ADJS': ['mean'],
'COMP': ['mean'],
'VERB': ['mean'],
'INFN': ['mean'],
'PRTF': ['mean'],
'PRTS': ['mean'],
'GRND': ['mean'],
'NUMR': ['mean'],
'ADVB': ['mean'],
'NPRO': ['mean'],
'PRED': ['mean'],
'PREP': ['mean'],
'CONJ': ['mean'],
'PRCL': ['mean'],
'INTJ': ['mean'],
'predict': ['sum']
})
group.columns = columns[4:]
group.reset_index(inplace=True)
data = pd.merge(group,
authors,
on='idauthor',
how='left',
suffixes=('', '_author'))
if mode == 'test':
data['predict'] = data['predict'].astype(int)
data = pd.merge(data,
authors,
left_on='predict',
right_on='idauthor',
how='left',
suffixes=('', '_predict'))
return data
def word2token(self, s):
t_start = timer()
env = Environment()
bgm_columns = env.bgm_columns_list(mode=1)
n_shift = 5
a_result=np.zeros(len(bgm_columns)+n_shift)
a_result[0] = len(s)
a_result[1] = self.s_encode(s[-2:]) # ts2
a_result[2] = self.s_encode(s[-3:]) # ts3
a_result[3] = self.s_encode(s[2:]) # tp2
a_result[4] = self.s_encode(s[3:]) # tp3
t_end = timer()
#env.debug(1, ['WordEncoder', 'word2token', '%s without bgm takes %s sec.' % (s, env.job_time(t_start, t_end))])
#t_start = timer()
di_letters = env.di_bgm_byletters
#print(di_letters)
di_word = {}
for n_l in range(0, len(s) - 1):
n_l2 = n_l + 1
di_n = di_letters.get('%s%s' % (s[n_l], s[n_l2]))
#print('%s%s' % (s[n_l], s[n_l2]),di_n)
if di_n is not None:
#print(di_n)
a_result[di_n + n_shift] = 1
t_end = timer()
#env.debug(1, ['WordEncoder', 'word2token', '%s takes %s sec.' % (s, env.job_time(t_start, t_end))])
return a_result
def model_train(self):
env = Environment()
data = self.stat()
t_start = timer()
y, X = self.model_prepare_data(data)
seed = 241
scoring = 'accuracy'
n_splits = 4
frac_test_size = 0.25
#Cross-validation
kf = KFold(n_splits=n_splits, shuffle=True, random_state=seed)
#clf = DecisionTreeClassifier(criterion='gini', random_state=seed)
#clf = GradientBoostingClassifier(n_estimators=50)
model = xgb.XGBClassifier(n_estimators=400,
max_depth=24,
colsample=1,
subsample=1,
seed=seed)
cv_scores = cross_val_score(model, X, y, cv=kf)
#eval
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=frac_test_size, random_state=seed)
eval_set = [(X_train, y_train), (X_test, y_test)]
#print(eval_set)
f_eval = 'merror'
# f_eval = 'mlogloss'
model.fit(X_train,
y_train,
eval_metric=f_eval,
eval_set=eval_set,
verbose=False,
early_stopping_rounds=10)
ev_scores = model.evals_result()
cv_mean = np.array(cv_scores.mean())
#ev_mean = np.array(ev_scores['validation_0']['mlogloss']).mean()
ev_mean = np.array(ev_scores['validation_0'][f_eval]).mean()
#Посмотрим важность признаков в модели
#print(model.feature_importances_)
xgb.plot_importance(model)
#plt.bar(range(len(model.feature_importances_)), model.feature_importances_)
plt.show()
#Обучаем модель на всех данных
model.fit(X, y, verbose=False)
#Сохраняем модель на диск
pickle.dump(model, open(env.filename_model_texts(), 'wb'))
#print('CV', cv_scores, 'EV', ev_scores)
print('Cross-validation: mean', cv_mean, 'eval_set mean', ev_mean)
return model
def visit_CombinatorNode(self, node):
env = Environment()
for parameter in node.parameters():
env.add(str(parameter))
d = env.index
self.visit('E', node.body(), env = env)
self.code.Update(d)
self.code.Pop(d)
self.code.Unwind()
self.symtab[node.name()].code = self.code.clone()
self.code.clear()
def visit_CombinatorNode(self, node):
env = Environment()
for parameter in node.parameters():
env.add(str(parameter))
d = env.index
self.visit('E', node.body(), env=env)
self.code.Update(d)
self.code.Pop(d)
self.code.Unwind()
self.symtab[node.name()].code = self.code.clone()
self.code.clear()
def model_predict(self, df, b_retrain=False):
env = Environment()
y, X = self.model_prepare_data(df, mode='test')
if b_retrain:
model = self.model_train(
) #Если хотим для кажжого теста вновь тренировать модель
else:
#Загружаем ранее тренированную модель с диска
model = pickle.load(open(env.filename_model_texts(), 'rb'))
#Предсказываем
y = model.predict(X)
return y
def pos_word_by_ml(self, awords):
env = Environment()
enc = Word_Encoder()
file_model = env.filename_model_tree()
clf = pickle.load(open(file_model, 'rb'))
a_predict = np.array([enc.word2token('')])
for word in awords:
a_padd = [enc.word2token(word)]
#print(word, a_padd)
a_predict = np.append(a_predict, a_padd, axis=0)
a_predict = a_predict[1:]
#print(a_predict[0, 100])
predictions = clf.predict(a_predict[:, 0:])
return (predictions[0:])
def model_prepare_data(self, df, mode='train'):
env = Environment()
data = df.copy()
data.drop(columns=['file', 'idchunk', 'predict'], inplace=True)
columns = data.columns
#idstat,idtext,idchunk,idauthor,author,name,file,words_all,words_chunk,sentences_all,sentence_mean,words_uniq,uniq_per_words,NOUN,ADJF,ADJS,COMP,VERB,INFN,PRTF,PRTS,GRND,NUMR,ADVB,NPRO,PRED,PREP,CONJ,PRCL,INTJ
columns2drop = [
'idtext', 'idauthor', 'author', 'name', 'sentences_text',
'words_text', 'sentences_chunk', 'words_chunk', 'words_uniq_chunk'
]
#New features
#Создадим новые статистические поля для помощи нашей модели
#data['words_uniq_per_sentense'] = data['words_uniq'] / data['sentences_all'] #кол-во уникальных слов/ кол-во предложений
#data['words_uniq_3k'] = data['words_uniq'] / 3000 # кол-во уникальных слов на 3 тыс. слов
#data['words_uniq_10k'] = data['words_uniq'] / 10000 #кол-во уникальных слов на 10 тыс. слов
y = None
if mode == 'train':
y = data['idauthor']
X = data.drop(columns=columns2drop)
#Add PCA features
n_components = 4
pca_cols2drop = [
'sentence_mean', 'uniq_per_sent_chunk', 'uniq_per_words_chunk'
]
if mode == 'train': #формируем матрицу признаков
pca_pos = PCA(n_components=n_components)
X_new = pca_pos.fit_transform(X.drop(columns=pca_cols2drop), y)
print(
'PCA ratio %s components quality: %s' %
(n_components,
round(np.sum(pca_pos.explained_variance_ratio_), 4)),
pca_pos.explained_variance_ratio_)
pickle.dump(pca_pos, open(env.filename_model_texts_pca(), 'wb'))
if mode == 'test': #Переводим признаки в пространство признаков на основе ранее созданной матрицы
pca_pos = pickle.load(open(env.filename_model_texts_pca(), 'rb'))
X_new = pca_pos.transform(X.drop(columns=pca_cols2drop))
for i in range(0, n_components):
X['pca_%s' % i] = X_new[:, i]
return y, X
def main():
parser = argparse.ArgumentParser(description="RL exercise.")
ADD = parser.add_argument
ADD('-e',
'--environment',
default='CartPole-v1',
help="Name of the OpenAI gym environment to train on.")
ADD('-m',
'--numEpisodesPerEval',
type=int,
default=100,
help="Number of episodes per policy update iteration.")
ADD('-n',
'--numIterations',
type=int,
default=1000,
help="Number of policy updates.")
ADD('-r',
'--renderEvery',
type=int,
default=0,
help="Render every nth episode. 0 to disable.")
ADD('-l',
'--learningRate',
type=float,
default=0.1,
help="Learning rate of policy update.")
ADD('-z',
'--sigma',
type=float,
default=0.2,
help=
"Standard deviation of policy for continuous actions. Exploration noise."
)
ADD('-p',
'--populationSize',
type=int,
default=64,
help="Population size.")
args = parser.parse_args()
# Create the environment, the policy and the training algorithm.
env = Environment(args.environment, args.numEpisodesPerEval,
args.renderEvery)
policy = DiscretePolicy(env)
algo = EvolutionStrategies(policy,
populationSize=args.populationSize,
sigma=args.sigma,
learnRate=args.learningRate)
# Train the policy.
algo.trainPolicy(policy, env, args.numIterations)
def stat(self):
env = Environment()
data = pd.DataFrame()
file_stat = env.filename_results_csv()
try:
data = pd.read_csv(file_stat, index_col='idstat', encoding='utf-8')
except:
env.debug(1, ['Failed to read stat file:', file_stat])
else:
env.debug(1, ['Read stat file OK:', file_stat])
#print(data)
return data
def tokenz(self):
env = Environment()
df_tokenz = pd.DataFrame()
file_tokenz = env.filename_tokenz_csv()
try:
df_tokenz = pd.read_csv(file_tokenz,
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(1, ['Failed to read tokenz file:', file_tokenz])
else:
env.debug(1, ['Read tokenz OK:', file_tokenz])
return df_tokenz
def authors(self, mode=0):
env = Environment()
df = pd.DataFrame()
filename = env.filename_authors_csv()
try:
df = pd.read_csv(filename, index_col='idauthor', encoding='utf-8')
except:
env.debug(1, ['Failed to load authors CSV file', filename])
else:
env.debug(1, ['Load authors CSV file', filename])
if mode == 1:
return df.to_dict().get('name')
else:
return df
def corpus_xml2txt(self, num=1, persistent=True):
result = True
env = Environment()
file_xml = env.filename_corpus_xml(num)
try:
tree = ET.ElementTree(file=file_xml)
except:
env.debug(1, ['Failed to load XML:', file_xml])
result = False
else:
file_txt = env.filename_corpus_txt(num)
file = open(file_txt, mode='w')
for elem in tree.iter('source'):
# print(elem.text, elem.tag, elem.attrib)
file.write(elem.text)
file.write(' ')
file.close()
env.debug(1, ['Write corpus file to TXT:', file_txt])
return result
def grammemes(self, mode=0):
env = Environment()
dfgram = pd.DataFrame()
filename_gram = env.filename_grammemes_csv()
try:
dfgram = pd.read_csv(filename_gram,
index_col='idgram',
encoding='utf-8')
except:
env.debug(1, ['Failed to load grammemes CSV file', filename_gram])
else:
env.debug(1, ['Load grammemes CSV file', filename_gram])
if mode == 1:
return dfgram.to_dict().get('name')
else:
return dfgram
def main():
parser = argparse.ArgumentParser(description="RL exercise.")
ADD = parser.add_argument
ADD('-e',
'--environment',
default='CartPole-v1',
help="Name of the OpenAI gym environment to train on.")
ADD('-m',
'--numEpisodesPerEval',
type=int,
default=100,
help="Number of episodes per policy update iteration.")
ADD('-n',
'--numIterations',
type=int,
default=1000,
help="Number of policy updates.")
ADD('-l',
'--learningRate',
type=float,
default=0.1,
help="Learning rate of policy update.")
ADD('-g',
'--gamma',
type=float,
default=0.99,
help="Rewards discount factor.")
ADD('-r',
'--renderEvery',
type=int,
default=0,
help="Render every nth episode. 0 to disable.")
args = parser.parse_args()
# Create the environment, the policy and the training algorithm.
env = Environment(args.environment, args.numEpisodesPerEval,
args.renderEvery)
policy = DiscretePolicy(env)
algo = ReinforceAlgorithm(policy,
gamma=args.gamma,
learnRate=args.learningRate)
# Train the policy.
algo.trainPolicy(policy, env, args.numIterations)
def main():
pd.set_option("display.max_columns", 100)
pd.set_option('display.width', 1000)
#Служебные классы
env = Environment()
c = OpenCorpus()
t = POSTagger()
a = mlAnalyzer()
enc = Word_Encoder()
g = pd.DataFrame()
g = c.grammemes()
dg = c.grammemes(
mode=1
) #Справочник Части речи mode = 1 возвращает в виде словаря python
da = c.authors(mode=1) #Справочник - авторы
#Пример обработки текстов из texts_train и добавления статистической информации в results
#a_texts_train = [1, 16]
#a_texts_train = [48]
#for i in a_texts_train:
# a.process_from_texts_file([i])
#Пример визуализации статистической информации из results в 2-мерном пространстве
#a.vizualize2d()
#Пример визуализации статистичесокй информации о частях речи
#t.vizualize2d(n_frac = 0.001)
#Предсказание автора текста из text_test
#[0, 1, 2, 3, 4]) #предсказать все тексты - долго
text2predict = [3]
y = a.predict(text2predict) #предсказать - указать номер текста
j = 0
for i in y:
print('idtext=%s' % text2predict[j], da.get(i))
j = j + 1
def tokenz_create_stat(self, dftokenz=pd.DataFrame(), n_frac=1):
env = Environment()
enc = Word_Encoder()
di_letters = Environment.di_bgm_byletters
bgm_columns = env.bgm_columns_list(mode=1)
t_start = timer()
if dftokenz.empty:
dftokenz = self.tokenz()
if n_frac < 1:
dftokenz = dftokenz.sample(frac=n_frac)
env.debug(1, [
'POStagger', 'create_stat',
'Collecting statistic START %s words' % dftokenz.shape[0]
])
di_tokenz_stat = (dftokenz.count()).to_dict()
di_tokenz_res = {}
#print('di_letters', di_letters)
print('di_tokenz_stat', di_tokenz_stat)
bgm_astat = [['init', 0]]
bgm_index = []
for key in di_letters:
di_n = di_letters.get(key)
column_stat = di_tokenz_stat.get(bgm_columns[di_n])
#di_tokenz_res[key] = column_stat
bgm_astat.append([key, column_stat])
bgm_index.append(di_n)
bgm_astat = bgm_astat[1:]
print('column stat', bgm_astat)
df_bgm_stat = pd.DataFrame(data=bgm_astat,
columns=['bigram', 'counts'],
index=bgm_index)
df_bgm_stat.index.name = 'idbigram'
df_bgm_stat = df_bgm_stat.sort_values(by=['counts'], ascending=False)
print('bgm_stat\n', df_bgm_stat)
df_bgm_stat.to_csv(env.filename_stat_bigram_letters_csv(),
encoding='utf-8')
def vizualize2d(self, n_frac=0.01, b_annotations=False):
n_components = 2
env = Environment()
c = OpenCorpus()
di_g = c.grammemes(mode=1)
data = self.tokenz().sample(frac=n_frac)
data = data.fillna(0)
#print(data['idgram'].shape)
#print(data.index.shape)
tdf = pd.DataFrame(index=data.index)
tdf['idgram'] = data['idgram']
tdf['gram'] = data['gram']
tdf['word'] = data['word']
#print(tdf)
drop_columns = [
'word', 'gram', 's_suffix2', 's_suffix3', 's_prefix2', 's_prefix3',
'n_token'
] # , 'bgm_l_None'
# drop_columns.extend(['bgm_l_%s' % (i) for i in range(1, env.bgm_columns_max()) if 'bgm_l_%s' % (i) not in bgm_columns])
env.debug(
1,
['POStagger', 'visualize2D',
'Drop colums: %s' % (drop_columns)])
data = data.drop(columns=drop_columns, axis=1)
values = data.values
X = values[:, 1:]
y = values[:, 0]
#print(data.head,X, y)
#return 0
#Scalers
sc = StandardScaler()
min_max_scaler = preprocessing.MinMaxScaler()
max_abs_scaler = preprocessing.MaxAbsScaler()
#X = sc.fit_transform(X)
#PCA
b_pca = False
b_sne = True
if b_pca:
model = PCA(n_components=n_components)
if b_sne:
model = MDS(n_components=n_components) #TSNE
X_new = model.fit_transform(X, y)
if b_pca:
print('PCA ratio', n_components, 'components',
model.explained_variance_ratio_)
#X_new = sc.fit_transform(X_new)
#X_new = preprocessing.scale(X_new)
if b_pca:
X_new = max_abs_scaler.fit_transform(X_new)
#return 0
#tdf = pd.DataFrame(data=X_new, columns=['PC1', 'PC2'], index=data.index)
tdf['PC1'] = X_new[:, 0]
tdf['PC2'] = X_new[:, 1]
#finalDf = pd.concat([tdf, data[['idgram']]], axis=1)
df_groups = tdf.groupby('idgram').count()
#print(df_groups)
#return 0
tdf['counts'] = 0
for index, serie in tdf.iterrows():
n_idgram = tdf.at[index, 'idgram']
tdf.at[index,
'counts'] = df_groups[df_groups.index == n_idgram]['gram']
tdf = tdf.sort_values(by=['counts'], ascending=False)
#print(tdf)
#Draw
i = 0
N = df_groups.shape[0]
s_title = ''
if b_pca:
s_title = '2 component PCA. Точность %s' % (round(
sum(float(i) for i in model.explained_variance_ratio_), 2))
if b_sne:
s_title = 't-SNE'
#Plotly
if False: #Plotly
py.sign_in('shashmaxus', 'AdfwTulrOoV3cSlbZT3B')
c = [
'hsl(' + str(h) + ',50%' + ',50%)'
for h in np.linspace(0, 360, N)
]
data_trace = []
for index, row in df_groups.iterrows():
#print(index)
df_trace = tdf[tdf['idgram'] == index]
#print(df_trace)
g_trace = go.Scatter(
x=df_trace['PC1'].values,
y=df_trace['PC2'].values,
name=df_trace['gram'].values[0],
mode='markers', #'markers+text'
marker=dict(
size=8,
color=i, #c[i]
opacity=0.8,
colorscale='Viridis'),
text=df_trace['word'],
textfont=dict(family='sans serif', size=12))
data_trace.append(g_trace)
i += 1
layout = go.Layout(
title=s_title_pca,
xaxis=dict(
title=('Component 1. Вклад %s' %
(round(pca.explained_variance_ratio_[0], 2)))),
yaxis=dict(
title=('Component 2. Вклад %s' %
(round(pca.explained_variance_ratio_[1], 2)))))
fig2 = go.Figure(data=data_trace, layout=layout)
py.image.save_as(fig2,
filename='c:/prj/mlivos_data/temp/Words2.png')
#Bokeh
if True:
palette = d3['Category20'][len(tdf['gram'].unique())]
#palette = all_palettes['Category20'][len(tdf['gram'].unique())]
#palette = Viridis256[len(tdf['gram'].unique())]
#palette = Viridis256
color_map = CategoricalColorMapper(factors=tdf['gram'].unique(),
palette=palette)
#print(mapper)
fig = figure(title=s_title, toolbar_location=None)
source = ColumnDataSource(tdf[['gram', 'PC1', 'PC2']])
fig.scatter(x='PC1',
y='PC2',
size=12,
color={
'field': 'gram',
'transform': color_map
},
legend='gram',
source=source)
show(fig)
export_png(fig, filename="c:/prj/mlivos_data/temp/PCA.png")
return 0
def predict(self, aidtext, b_makestat=False):
env = Environment()
# Открываем файл со статистикой по тестовым текстам
df_stat = pd.read_csv(
env.filename_stat_test_csv(), index_col='idstat',
encoding='utf-8') # Статистика по тстовым текстам
df_texts = pd.read_csv(env.filename_predict_csv(),
index_col='idtext',
encoding='utf-8') # Реестр текстов
mask = df_texts.index.isin(aidtext)
df_texts = df_texts[mask]
columns = ['idtext', 'idchunk', 'idauthor', 'author', 'name', 'file', \
'sentences_text', 'words_text','sentence_mean', \
'sentences_chunk', 'words_chunk',
'words_uniq_chunk','uniq_per_sent_chunk','uniq_per_words_chunk', \
'NOUN','ADJF','ADJS','COMP','VERB','INFN','PRTF','PRTS','GRND','NUMR',\
'ADVB','NPRO','PRED','PREP','CONJ','PRCL','INTJ', 'predict']
y_result = []
#Если необходимо подготовить статистику по тестовым текстам
if b_makestat:
for index, row in df_texts.iterrows(
): # Для каждого текста, который надо обработать
file_txt = df_texts.at[index, 'filename']
# Read text file
env.debug(1,
['Analyzer', 'predict', 'START file TXT:', file_txt])
t_start = timer()
file = codecs.open(file_txt, "r", "utf_8_sig")
text = file.read().strip()
file.close()
# Автор в тестовой выборке вообще говоря нет
idauthor = df_texts.at[index, 'idauthor'] # Автор
#idauthor = 0
name = df_texts.at[index, 'name'] # Название
# Собственно обработка текста
df_add = self.analyze_text(
columns, text, index, idauthor, name,
file_txt) # Analyze text, get Series
#print(df_add)
df_add.reset_index(drop=True, inplace=True)
df_stat = df_stat.append(
df_add, ignore_index=True) #Добавляем к файлу результатов
df_stat.reset_index(drop=True, inplace=True)
df_stat.index.name = 'idstat'
t_end = timer()
env.debug(1, [
'END file TXT:', file_txt, 'time:',
env.job_time(t_start, t_end)
])
#df_stat теперь содержит информацию о всех тестовых текстах, которые хотели обработать
#Указываем верный тип для целочисленных колонок
int_cols = [
'idtext', 'idchunk', 'idauthor', 'sentences_text',
'words_text', 'sentences_chunk', 'words_chunk',
'words_uniq_chunk'
]
for col in int_cols:
df_stat[col] = df_stat[col].astype(int)
# Сохраняем результат на диск
df_stat.to_csv(env.filename_stat_test_csv(), encoding='utf-8')
#Статистика готова
# Открываем файл со статистикой по тестовым текстам
df_stat = pd.read_csv(
env.filename_stat_test_csv(), index_col='idstat',
encoding='utf-8') # Статистика по тстовым текстам
#mask = df_stat.index.isin(aidtext)
#df_stat2predict = df_stat[mask]
#Предсказываем авторов
y_res = self.model_predict(df_stat.loc[aidtext])
#print(y_res)
df_stat.loc[aidtext, 'predict'] = y_res.astype(int)
#print(df_stat)
#y_result.append(y_res[0])
#Сохраняем измененный файл с предсказаниями
df_stat.to_csv(env.filename_stat_test_csv(), encoding='utf-8')
return y_res #Возвращаем предсказания
def vizualize2d(self, mode='train'):
n_components = 2
env = Environment()
data = self.get_texts_stat(mode=mode)
columns = data.columns
#print(data)
#print(columns)
columns2drop = [
'idtext', 'idauthor', 'author', 'name', 'sentences_text',
'words_text', 'sentence_mean', 'sentences_chunk', 'words_chunk',
'words_uniq_chunk', 'uniq_per_sent_chunk', 'predict', 'shortname',
'name_author'
]
y = data['idauthor'].values
X = data.drop(columns=columns2drop).values
#print(y, X)
#return 0
#print(data)
#print(X, y)
pca = PCA(n_components=n_components)
#pca = TSNE(n_components=2)
X_new = pca.fit_transform(X, y)
print('PCA ratio 2 components', pca.explained_variance_ratio_)
#print('components', pca.components_)
#print(X_new)
tdf = pd.DataFrame(data=X_new, columns=['PC1', 'PC2'])
finalDf = pd.concat([tdf, data[['idauthor', 'name', 'shortname']]],
axis=1)
print('dataframe ', finalDf)
mpl.style.use('default')
rcParams['font.family'] = 'sans-serif'
rcParams['font.sans-serif'] = ['Tahoma']
fig = plt.figure(figsize=(8, 8))
ax = fig.add_subplot(1, 1, 1)
ax.set_xlabel('Component 1. Вклад ' +
str(round(pca.explained_variance_ratio_[0], 2)),
fontsize=12)
ax.set_ylabel('Component 2. Вклад ' +
str(round(pca.explained_variance_ratio_[1], 2)),
fontsize=12)
ax.set_title(
'2 component PCA. Точность ' +
str(round(sum(float(i)
for i in pca.explained_variance_ratio_), 2)),
fontsize=12)
targets = data.idauthor.unique()
print(targets)
legends = data.shortname.unique()
print(legends)
#print(targets)
#colors = ['r', 'g', 'b']
#colors = "bgcmykw" #without r
#colors = ['#1f77b4', '#ff7f0e', '#2ca02c', '#d62728',
# '#9467bd', '#8c564b', '#e377c2', '#7f7f7f',
# '#bcbd22', '#17becf']
colors = [
"#1f77b4", "#aec7e8", "#ff7f0e", "#ffbb78", "#2ca02c", "#98df8a",
"#d62728", "#ff9896", "#9467bd", "#c5b0d5", "#8c564b", "#c49c94",
"#e377c2", "#f7b6d2", "#7f7f7f", "#c7c7c7", "#bcbd22", "#dbdb8d",
"#17becf", "#9edae5"
]
for target in targets:
indicesToKeep = finalDf['idauthor'] == target
ax.scatter(finalDf.loc[indicesToKeep, 'PC1'],
finalDf.loc[indicesToKeep, 'PC2'],
c=colors[target],
s=50)
for index, row in finalDf.iterrows():
ax.annotate(
finalDf.at[index, 'name'],
xy=(finalDf.at[index, 'PC1'], finalDf.at[index, 'PC2']),
# xytext=(0.05, 0.05),
fontsize=8)
ax.legend(legends)
ax.grid()
plt.show()
def vocabulary(self):
env = Environment()
file_voc = env.filename_vocabulary_csv() #from vocabulary file
file_dict = env.filename_dict_csv() #from dictionary file
try:
df_voc = pd.read_csv(file_voc,
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(1, ['Failed to read vocabulary file:', file_voc])
else:
env.debug(1, ['Read vocabulary OK:', file_voc])
try:
df_dict = pd.read_csv(file_dict,
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(1, ['Failed to read dictionary file:', file_dict])
else:
env.debug(1, ['Read dictionary OK:', file_dict])
#Concat
df_res = pd.concat([df_voc, df_dict])
df_res = df_res.drop_duplicates()
#Apply patch words
df_patch = pd.read_csv(env.filename_vocabulary_patch_csv(),
index_col='idcorpus',
encoding='utf-8')
df_res = df_res.drop(df_res[df_res['word'].isin(
df_patch['word'])].index,
axis=0)
df_res = pd.concat([df_res, df_patch])
#print(df_res[df_res['word'].isin(df_patch['word'])])
df_res = df_res.reset_index(drop=True)
df_res.index.name = 'idcorpus'
#print(df_res)
return df_res
def make_report(self):
env = Environment()
a = mlAnalyzer()
#template = Template('Hello {{ name }}!')
#print(template.render(name=u'Вася'))
jenv = jinjaEnvironment(loader = FileSystemLoader(env.path_templates()))
#print(jenv.loader)
template = jenv.get_template("report_global.tpl.html")
data = a.get_texts_stat() #Статистика по текстам из файла обучающей выборки
test = a.get_texts_stat(mode='test') #Статистика по текстам тестовой выборки
#print(data)
#test['predict'] = test['predict'].astype(int)
test['validation'] = 0
test.loc[test.idauthor == test.predict,'validation'] = 1
print(data)
print(test)
#Summary stat
group = pd.merge(data, test, on='idauthor', how='left', suffixes=('', '_test'))
print(group)
group = group.groupby(['idauthor', 'name_author'], as_index=False).agg({'idtext' : ['nunique'],
'words_chunk' : ['sum'],
'name_test': ['nunique'],
'words_chunk_test': ['sum'],
'validation' : ['mean']
})
print(group)
group.drop(['idauthor'], axis=1, inplace=True)
group.sort_values('name_author', inplace=True)
#Переименовать колонки в русскоязычные
group.columns = ['Писатель',
'Кол-во текстов для обучения',
'Объём текстов для обучения (кол-во слов)',
'Кол-во текстов для проверки',
'Объём текстов для проверки (кол-во слов)',
'Точность определения'
]
n_accuracy = group['Точность определения'].mean()
#Целые числа показываем без дробной части
int_cols = ['Кол-во текстов для обучения',
'Объём текстов для обучения (кол-во слов)',
'Кол-во текстов для проверки',
'Объём текстов для проверки (кол-во слов)']
for col in int_cols:
group[col] = group[col].astype(int)
group.reset_index(drop = True, inplace = True)
s = group.style.set_properties(**{'text-align': 'right'})
group.fillna('', inplace = True)
s.hide_index().render()
#Training stat
group_train = data.groupby(['author'], as_index=False).agg({'idauthor' : ['count'],
'sentences_text' : ['sum'],
'words_text' : ['sum'],
'sentence_mean': ['mean'],
'name': [lambda col: '<br />'.join(col)],
})
group_train.reset_index(drop = True, inplace = True)
s_train = group_train.style.set_properties(**{'text-align': 'right'})
group_train.fillna('', inplace=True)
group_train.columns = ['Писатель',
'Кол-во текстов',
'Кол-во предложений',
'Кол-во слов',
'Средняя длина предложения',
'Произведения'
]
n_train = group_train['Кол-во текстов'].sum()
s_train.hide_index().render()
# Testing stat
group_test = test.groupby(['author'], as_index=False).agg({'idauthor': ['count'],
'sentences_text': ['sum'],
'words_text': ['sum'],
'sentence_mean': ['mean'],
'name': [lambda col: '<br />'.join(col)],
'validation': ['mean'],
'shortname_predict': [lambda col: '<br />'.join(col)],
})
group_test.reset_index(drop=True, inplace=True)
s_test = group_test.style.set_properties(**{'text-align': 'right'})
group_test.fillna('', inplace=True)
group_test.columns = ['Писатель',
'Кол-во текстов',
'Кол-во предложений',
'Кол-во слов',
'Средняя длина предложения',
'Произведения',
'Результат проверки',
'Определён автор',
]
n_test = group_test['Кол-во текстов'].sum()
s_test.hide_index().render()
template_vars = {"title": "Отчёт",
"detection_accuracy": '%s' % (round(n_accuracy,4)*100),
"train_texts_pivot_table_style_render" : s.render(),
"n_train_texts": round(n_train,0),
"train_texts_table_style_render": s_train.render(),
"n_test_texts": round(n_test,0),
"test_texts_table_style_render": s_test.render()
}
html_out = template.render(template_vars)
file = codecs.open(env.filename_global_report_html(), "w", "utf-8-sig")
file.write(html_out)
file.close()
#print(html_out)
return html_out
def test(self, n_min=1, n_max=1):
t_start = timer()
env = Environment()
df_test = pd.DataFrame()
for i in range(n_min, n_max + 1):
try:
dffile = pd.read_csv(env.filename_corpus_csv(i),
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(1, [
'POStagger', 'test', 'Failed to read corpus file:',
env.filename_corpus_csv(i)
])
else:
env.debug(1, [
'POStagger', 'test', 'Read OK:',
env.filename_corpus_csv(i)
])
if not dffile.empty:
df_test = df_test.append(dffile)
df_test = df_test.drop_duplicates()
df_test.columns = ['word', 'gram', 'idgram']
df_test = df_test.reset_index(drop=True)
df_test.index.name = 'idcorpus'
df_test['gram_valid'] = df_test['gram']
n_testsize = df_test.shape[0]
env.debug(1, ['POStagger', 'test', 'START %s words' % n_testsize])
df_test = self.pos(df_test)
print('Test result', df_test)
df_err = df_test[df_test['gram_valid'] != df_test['gram']]
print('Test errors:', df_err)
df_err.to_csv(env.filename_test_err_csv(), encoding='utf-8')
env.debug(1, [
'POStagger', 'test',
'test accuracy %s' % (1 - df_err.shape[0] / n_testsize)
])
t_end = timer()
env.debug(1, [
'POSTagger', 'test', 'test time:',
env.job_time(t_start, t_end), 'sec.'
])
def corpus_xml2csv(self, num=1, persistent=True):
env = Environment()
file_xml = env.filename_corpus_xml(num)
df_xml = pd.DataFrame()
df_gram = self.grammemes()
dgram = df_gram.to_dict().get('name')
try:
tree = ET.ElementTree(file=file_xml)
except:
env.debug(1, ['Failed to load XML:', file_xml])
else:
t_start = timer()
env.debug(1, ['CORPUS', 'XML to CSV:', file_xml])
for elem in tree.iter('token'):
#print(elem.tag, elem.attrib)
serie = pd.Series(data=[])
badd = False
s_text = elem.attrib.get('text')
serie[len(serie)] = s_text.lower()
for elem2 in elem.iter('g'):
#print(elem2.tag, elem2.attrib)
sgram = elem2.attrib.get('v')
sgram = sgram.upper()
if (df_gram[df_gram['name'].isin([sgram]) == True].size
) > 0:
serie[len(serie)] = sgram
serie[len(serie)] = int(df_gram.index[
df_gram['name'] == sgram].tolist()[0])
#serie[len(serie)] = list(dgram.keys())[list(dgram.values()).index(sgram)]
badd = True
break
#print(s)
if badd:
df_xml = df_xml.append(serie, ignore_index=True)
if not df_xml.empty:
df_xml = df_xml.drop_duplicates()
df_xml = df_xml.reset_index(drop=True)
df_xml.index.name = 'idcorpus'
df_xml.columns = ['word', 'gram', 'idgram']
df_xml = df_xml.astype({"idgram": int})
if persistent:
file_csv = env.filename_corpus_csv(num)
env.debug(1, ['Write corpus file to CSV:', file_csv])
df_xml.to_csv(file_csv, encoding='utf-8')
t_end = timer()
env.debug(1, [
'CORPUS', 'CSV written:', file_csv,
'takes %s sec.' % env.job_time(t_start, t_end)
])
return df_xml
def train(self,
df=pd.DataFrame(),
validation='eval',
n_splits=5,
b_smoketest=True,
n_frac=1):
env = Environment()
enc = Word_Encoder()
df_train = df
bgm_columns = env.bgm_columns_list(mode=1)
drop_columns = [
'word', 'gram', 's_suffix2', 's_suffix3', 's_prefix2', 's_prefix3',
'n_token'
] #, 'bgm_l_None'
#drop_columns.extend(['bgm_l_%s' % (i) for i in range(1, env.bgm_columns_max()) if 'bgm_l_%s' % (i) not in bgm_columns])
env.debug(1,
['POStagger', 'train',
'Drop colums: %s' % (drop_columns)])
if df_train.empty:
t_start = timer()
df_train = self.tokenz()
t_end = timer()
env.debug(1, [
'POSTagger', 'train', 'tokenz loaded:', 'time:',
env.job_time(t_start, t_end)
])
env.debug(1, [
'POStagger', 'train',
'All tokenz set shape %s' % df_train.shape[0]
])
t_start = timer()
env.debug(1, ['POStagger', 'train', 'Learning: START'])
if n_frac < 1:
df_train = df_train.sample(frac=n_frac)
env.debug(1, [
'POStagger', 'train',
'Training tokenz set shape %s' % df_train.shape[0]
])
#print(df_train.shape)
#df_train2 = df_train[bgm_columns]
#print(df_train2.shape)
#df_train2 = df_train2.astype({"idgram": int})
df_train = df_train.drop(columns=drop_columns, axis=1)
env.debug(
1, ['POStagger',
'Train colums: %s' % (df_train.columns.tolist())])
#print(df_train.columns)
#df_train = df_train.drop_duplicates() #slow-slow
#print(df_train.head())
df_train = df_train.fillna(0)
file_x = env.filename_xtrain_csv()
df_train.to_csv(file_x, encoding='utf-8')
env.debug(1, ['POStagger', 'train', 'Save X', file_x])
y = df_train['idgram'].values
df_train.drop(columns=['idgram'], inplace=True)
X = df_train.values
#array = df_train.values
#print(df_train)
#X = array[:, 1:]
#Y = array[:, 0]
#print(X, Y)
#validation_size = 0.20
seed = 241
frac_test_size = 0.2
sc = StandardScaler()
#Y_sc = sc.fit_transform(Y)
t2_start = timer()
if validation == 'cv': #Need cross-validation
scoring = 'accuracy'
# scoring = 'f1_samples'
kf = KFold(n_splits=n_splits, shuffle=True, random_state=seed)
if True: #Decision tree
env.debug(1, ['Tree cross-validation'])
# clf = DecisionTreeClassifier(criterion='gini', random_state=seed) # 0.79
# clf = KNeighborsClassifier(n_neighbors=230)
model = DecisionTreeClassifier(criterion='entropy',
random_state=seed) # 0.81
env.debug(
1, ['Calculate cross_val_score. Splits=%s' % (n_splits)])
scores = cross_val_score(model, X, y, cv=kf)
print('DTree scores:', scores.mean(), 'raw', scores)
if False: #Logistic regression
env.debug(1, ['LGR cross-validation'])
n_Cs = [0.01]
X = array[:, 5:]
X_sc = sc.fit_transform(X)
Y = df_train['idgram'].values
Y[Y > 0] = 1
print(X_sc, Y)
for n_c in n_Cs:
#clf = LogisticRegression(penalty='l2', solver='saga', C=n_c, multi_class='multinomial')
clf = LogisticRegression(penalty='l2',
solver='liblinear',
C=n_c)
# clf = SVC(kernel='linear', C=10000, random_state=241)
# clf = SVC(kernel='linear', C=0.01, random_state=seed)
# clf = SVC(random_state=seed)
# clf = Perceptron()
env.debug(1, [
'Calculate cross_val_score. Splits=%s C=%s' %
(n_splits, n_c)
])
scores = cross_val_score(clf, X_sc, Y, cv=kf)
print(scores)
if False: #GBM, RandomForest
env.debug(1, ['GBM cross-validation'])
asteps = [20] #GBM
#asteps=[100] #RandomForest
for i in asteps:
#clf = RandomForestClassifier(n_estimators=i)
clf = GradientBoostingClassifier(
n_estimators=i, max_depth=8) #, max_features='sqrt'
env.debug(1, [
'Calculate cross_val_score. Splits=%s Estimators=%s' %
(n_splits, i)
])
scores = cross_val_score(clf, X, Y, cv=kf)
print(scores)
if validation == 'eval':
# eval
model = xgb.XGBClassifier(n_estimators=140,
max_depth=16,
colsample=1,
subsample=0.5,
seed=seed)
X_train, X_test, y_train, y_test = train_test_split(
X,
y,
test_size=frac_test_size,
random_state=seed,
shuffle=True)
eval_set = [(X_train, y_train), (X_test, y_test)]
# print(eval_set)
f_eval = 'merror'
# f_eval = 'mlogloss'
model.fit(X_train,
y_train,
eval_metric=f_eval,
eval_set=eval_set,
verbose=False,
early_stopping_rounds=20)
ev_scores = model.evals_result()
ev_mean = np.array(ev_scores['validation_0'][f_eval]).mean()
#print(model.feature_importances_)
print(ev_mean, ev_scores)
xgb.plot_importance(model)
plt.show()
t2_end = timer()
t_end = timer()
env.debug(1, ['CV completed:', 'time:', env.job_time(t_start, t_end)])
if validation == 'cv':
#Training на всех данных
X_train, y_train = X, y
# model = SVC()
# model= DecisionTreeClassifier() #79
# model= LinearDiscriminantAnalysis() #47
# model=LogisticRegression() #48
# model = KNeighborsClassifier(n_neighbors=200) #48
# model = GaussianNB() #43
#print('Fit...')
#print('Validate...')
# predictions = model.predict(X_validation)
# print(accuracy_score(Y_validation, predictions))
# print(confusion_matrix(Y_validation, predictions))
# print(classification_report(Y_validation, predictions))
t_start = timer()
env.debug(1, ['Training: START'])
model.fit(X_train, y_train)
t_end = timer()
env.debug(1, ['Training: END', env.job_time(t_start, t_end)])
pickle.dump(sc, open(env.filename_scaler(), 'wb'))
pickle.dump(model, open(env.filename_model_tree(), 'wb'))
# Smoke test
if b_smoketest:
X_smoke_predict = [
'съеште', 'ещё', 'этих', 'мягких', 'французских', 'булок'
]
a_smoke = np.array(
[enc.word2token(elem) for elem in X_smoke_predict])
y_predictions = model.predict(a_smoke[:, 0:])
y_predictions_proba = model.predict(a_smoke[:, 0:])
#print(y_predictions)
print('Prediction', list(zip(X_smoke_predict, y_predictions)))
print('Proba', list(zip(X_smoke_predict, y_predictions_proba)))
return model
def dict_xml2csv(self, persistent=True, lines=10000):
t_start = timer()
env = Environment()
dfgram = self.grammemes()
filename_dict = env.filename_dict_xml()
dfcols = ['word', 'gram', 'idgram']
df_xml = pd.DataFrame(columns=dfcols)
env.debug(
1, ['CORPUS', 'Start to load dictionary from XML:', filename_dict])
try:
fp = io.open(filename_dict, mode="r", encoding="utf-8")
except:
env.debug(1, [
'CORPUS', 'Failed to open dictionary file XML:', filename_dict
])
else:
number_lines = sum(1 for line in fp)
fp.seek(0)
t_end = timer()
env.debug(1, [
'CORPUS', 'File opened:', 'lines',
'%s' % number_lines, 'time:',
env.job_time(t_start, t_end)
])
t_start = timer()
step = number_lines // lines
env.debug(1, [
'CORPUS', 'Read dictionary:', filename_dict,
'lines: %s step %s' % (lines, step)
])
n_line = 0
for i in range(0, number_lines):
line = fp.readline()
#print(line[5:10])
if (line[5:10] == 'lemma') and (n_line == 0):
#print(line)
tree = ET.fromstring(line)
for elem in tree.iter('l'):
s_word = elem.attrib.get('t')
gram = ['', 0]
j = 0
for elem2 in elem.iter('g'):
gram[j] = elem2.attrib.get('v')
break
gram[1] = int(dfgram.index[dfgram['name'] ==
gram[0]].tolist()[0])
#print(s_word,gram)
s = pd.Series(data=[s_word, gram[0], gram[1]],
index=dfcols)
df_xml = df_xml.append(s, ignore_index=True)
n_line += 1
n_line += 1
if n_line >= step:
n_line = 0
fp.close()
df_xml.index.name = 'idcorpus'
t_end = timer()
env.debug(1, [
'CORPUS', 'Dictionary loaded:', 'time:',
env.job_time(t_start, t_end)
])
if persistent:
filename_csv = env.filename_dict_csv()
env.debug(1,
['CORPUS', 'Write dictionary to CSV:', filename_csv])
df_xml.to_csv(filename_csv, encoding='utf-8')
env.debug(1, ['CORPUS', 'Dictionary saved:', filename_csv])
return df_xml
def main():
pd.set_option("display.max_columns", 100)
pd.set_option('display.width', 1000)
env = Environment()
c = OpenCorpus()
t = POSTagger()
a = mlAnalyzer()
enc = Word_Encoder()
r = Reporter()
#c.dict_xml2csv(lines = 600000)
#c.grammemes_xml2csv()
#c.vocabulary_from_corpus(1,1000)
g = pd.DataFrame()
g = c.grammemes()
#dg = g.to_dict().get('name')
dg = c.grammemes(mode=1) #grammemes by id
da = c.authors(mode=1) # authors by id
#print(dg)
#print(p.head())
#for i in range(2015,3000):
# c.corpus_xml2csv(i)
#c.corpus_xml2csv(2)
#for i in range (125,150):
# c.corpus_xml2txt(i)
#print(c.vocabulary_from_corpus(1,2000).head())
#voc=c.vocabulary()
#print(voc.head())
#t.tokenize()
#print(t.tokenize(voc, n_frac=1))
#t.tokenz_create_stat()
#print(env.bgm_stat())
#print(t.tokenz())
#print(c.vocabulary())
#print(enc.word2token('паровоз'))
#print(enc.word2token('аз'))
#t.train(n_frac=0.8, validation='cv')
#t.train(n_frac=0.95, validation='eval')
#t.test(2000,2048)
#a.process_from_texts_file([49], mode='chunk_size')
#a.process_from_texts_file([58], max_words = 8000)
#arrt = [2, 45, 43, 44, 42, 40, 41, 46, 36, 37, 38, 34]
#arrt = [69]
#for i in range (51,95):
#for i in arrt:
#a.process_from_texts_file([i], max_words = 8000)
#t.vizualize2d(n_frac=0.01)
#nltk.download()
#a.vizualize2d()
#a.vizualize2d(mode='train')
#a.vizualize2d(mode='test')
#a.model_train()
#return 0
#y = a.predict([0, 1, 2, 3, 4])
#y = a.predict([0, 1, 2, 3, 4])
print(a.predict([16], b_makestat=True))
a.vizualize2d(mode='test')
#for i in y:
# print('idtext=%s' % i, da.get(i))
#text2predict = [11, 12, 13, 14, 15]
#y = a.predict(text2predict, b_makestat=True) # предсказать - указать номер текста
#j = 0
#for i in y:
# print('idtext=%s' % text2predict[j], 'Автор=%s (%s)' % (i, da.get(i)))
# j = j + 1
#predict=(t.pos_word_by_voc(['съеште', 'школа','господина','приехал',
# 'глокая','куздра','штеко','будланула','бокра','и','кудрячит','бокрёнка']))
X_predict = [
'съеште', 'школа', 'господина', 'приехал', 'глокая', 'куздра', 'штеко',
'будланула', 'бокра', 'и', 'кудрячит', 'бокрёнка', 'он', 'видел', 'их',
'семью', 'своими', 'глазами'
]
#X_predict=['символ']
y_predict = t.pos_word_by_ml(X_predict)
print([
'%s/%s' % (X_predict[i], dg.get(y_predict[i]))
for i in range(0, len(y_predict))
])
r.make_report()
def pos(self, df, mode_fast=True, use_cache=True):
env = Environment()
enc = Word_Encoder()
df_res = df
t_start = timer()
c = OpenCorpus()
g = c.grammemes()
dg = g.to_dict().get('name')
#Cache file
cache_columns = ['word', 'gram_ml', 'count']
file_cache = env.filename_mlcache_csv()
try:
df_cache = pd.read_csv(file_cache,
index_col='idcorpus',
encoding='utf-8')
except:
env.debug(
1,
['POSTagger', 'pos', 'Failed to read cache file:', file_cache])
df_cache = pd.DataFrame(columns=cache_columns)
else:
env.debug(1, ['POSTagger', 'pos', 'Read ML cache OK:', file_cache])
a_predict = np.array([enc.word2token('')])
#a_words = ['']
n_words = df_res.shape[0]
env.debug(1, [
'POStagger', 'pos',
'START Vocabulary prediction %s words' % n_words
])
a_words = df_res['word'].tolist()
a_ml_words = []
predictions_voc = self.pos_by_voc(a_words)
p_se = pd.Series(predictions_voc)
df_res['gram'] = p_se.values
df_res['gram_voc'] = p_se.values
df_res['gram_ml'] = ''
t_end = timer()
env.debug(1, [
'POStagger', 'pos',
'END Vocabulary prediction %s sec.' % env.job_time(t_start, t_end)
])
#print(predictions_voc)
if mode_fast:
#env.debug(1, ['POStagger', 'pos', 'START Fast mode vocabulary search. Words %s' % df.shape[0]])
df_ni_voc = df_res[df_res['gram_voc'] == '']
n_words = df_ni_voc.shape[0]
else:
df_ni_voc = df_res
#print('non-vocabulary',df_ni_voc)
if not df_ni_voc.empty:
env.debug(
1, ['POStagger', 'pos',
'START Encoding %s words' % n_words])
for index, serie in df_ni_voc.iterrows():
word = df_ni_voc.at[index, 'word']
#print(word)
a_padd = np.array([enc.word2token(word)])
a_predict = np.append(a_predict, a_padd, axis=0)
a_ml_words.append(word)
#print(a_words, a_predict)
a_predict = a_predict[1:, :]
#print(a_predict)
#print('ml_words',a_ml_words)
t_end = timer()
env.debug(1, [
'POStagger', 'pos',
'END Encoding %s words %s sec.' %
(n_words, env.job_time(t_start, t_end))
])
t_start = timer()
env.debug(1, ['POStagger', 'pos', 'START Model prediction'])
clf = pickle.load(open(env.filename_model_tree(), 'rb'))
predictions_ml = clf.predict(a_predict[:, 0:])
# print('ml', predictions_ml)
t_end = timer()
env.debug(1, [
'POStagger', 'pos',
'END Model prediction %s sec.' % env.job_time(t_start, t_end)
])
#print('ml_words_prediction',list(zip(a_ml_words,predictions_ml)))
t_start = timer()
i = 0
s_pvoc = ''
s_pml = ''
for index, row in df_res.iterrows():
word = df_res.at[index, 'word']
s_pvoc = df_res.at[index, 'gram_voc']
#s_pvoc = predictions_voc[i]
#print('s_pvoc', word, s_pvoc)
#df_res.at[index, 'gram_voc'] = s_pvoc
if s_pvoc == '':
if mode_fast:
try:
j = a_ml_words.index(word)
except:
pass
else:
s_pml = dg.get(predictions_ml[j])
#print(word,s_pml)
else:
s_pml = dg.get(predictions_ml[i])
df_res.at[index, 'gram_ml'] = s_pml
df_res.at[index, 'gram'] = s_pml
i = i + 1
t_end = timer()
env.debug(1, [
'POStagger', 'pos',
'ML predictions dataframe filled %s sec' %
env.job_time(t_start, t_end)
])
#print(df_res)
df_cache = pd.concat([
df_cache,
df_res[df_res.gram_ml != ''][['word', 'gram_ml', 'count']]
])
df_cache = df_cache.groupby(['word',
'gram_ml']).agg({'count': ['sum']})
df_cache.reset_index(inplace=True)
df_cache.index.name = 'idcorpus'
df_cache.columns = cache_columns
df_cache.sort_values(by=['count'], inplace=True, ascending=False)
#print(df_cache)
env.debug(1,
['POStagger', 'pos', 'Write ML cache to CSV:', file_cache])
df_cache.to_csv(file_cache, encoding='utf-8')
return df_res
def tokenize(self, dftokenz=pd.DataFrame(), persistent=True, n_frac=1):
env = Environment()
enc = Word_Encoder()
t_start = timer()
if dftokenz.empty:
dftokenz = self.tokenz()
if n_frac < 1:
dftokenz = dftokenz.sample(frac=n_frac)
env.debug(
1, ['Transforming to tokenz: START %s words' % dftokenz.shape[0]])
gmask = dftokenz.groupby(['gram'])
df_posstat = gmask.count()
df_posstat.to_csv(env.filename_stat_pos_tokenz_csv(), encoding='utf-8')
print('POSTagger', 'train dataset stat:\n', gmask.count())
fields = [
's_suffix2', 's_suffix3', 's_prefix2', 's_prefix3', 'n_token',
'n_len', 'n_tokens2', 'n_tokens3', 'n_tokenp2', 'n_tokenp3'
]
for field in fields:
val = 0.0
if field[0] == 's':
val = ''
dftokenz[field] = val
n_letters = 0
s_letters = env.list_rus_letters()
di_letters = env.di_bgm_byletters
#bgm_columns_i = env.bgm_columns_list(mode=0)
bgm_columns = env.bgm_columns_list(mode=1)
#print('bgm_columns', bgm_columns)
for column_name in bgm_columns:
dftokenz[column_name] = None
t_end = timer()
env.debug(1, [
'POStagger', 'Letters bigram columns added',
env.job_time(t_start, t_end)
])
#Form tokenz
t_start = timer()
for index, serie in dftokenz.iterrows():
# print (serie.values)
a_word = enc.s2token(index, serie)
i = 2
# print(a_word)
for field in fields:
dftokenz.at[index, field] = a_word[i]
# print(field, a_word[i])
i = i + 1
# print(dftokenz.loc[index])
#Letters bigram binaries
for n_l in range(0, len(a_word[0]) - 1):
n_l2 = n_l + 1
di_n = di_letters.get('%s%s' %
(a_word[0][n_l], a_word[0][n_l2]))
if di_n is not None:
#print(di_n)
#print(bgm_columns[di_n])
dftokenz.at[index, bgm_columns[di_n]] = 1
t_end = timer()
env.debug(
1,
['Transforming to tokenz: COMPLETE',
env.job_time(t_start, t_end)])
if persistent:
dftokenz.to_csv(env.filename_tokenz_csv(), encoding='utf-8')
env.debug(1, ['Tokenz written to CSV:', env.filename_tokenz_csv()])
return dftokenz
