def ru_words_test():
lang = 'russian'
if lang == 'russian':
reader = Reader('elsevier journal.pdf',
generateWordList=True) #, language='russian')
reader.loadFile()
words, lines, articles = reader.parseDocPages(
19, 21) #(14, 17)#0, 50)#()#2160, 2190)#0, 50)
for i in range(11):
print(i)
if i != 6:
print(' '.join(
[x['text'] for x in words if x['block_count'] == i]))
else:
for j in range(15):
print(j, ' sentence')
print(' '.join([
x['text'] for x in words
if x['block_count'] == i and x['sentence_count'] == j
]))
print("--=-")
else:
df_source = loadArticlesInEnglish()
def list_w_p(path, mass):
'''Return a list of work-power datas caculated from\
the .txt files within the path.\
Argument:\
mass: mass or surface area or volume of the sample'''
instances = []
for s in os.listdir(path):
# find the .txt files
if '.txt' in s:
f_path = os.path.join(path, s)
inst = rd.GV(filename=f_path, mass=mass)
instances.append(inst)
# the list for line objects in an axe
w_p = pd.DataFrame()
for counter, inst in enumerate(instances):
w_p = w_p.append(inst.work_power())
return w_p.reset_index(drop=True)
def lemmaEng():
lang = 'english'
#lang = 'russian'
partMatchesCounted = False #if true then custom, else rough precision recall
num = 8
if lang == 'russian':
reader = Reader('elsevier journal.pdf') #, language='russian')
reader.loadFile()
words, lines, articles = reader.parseDocPages() #2160, 2190)#0, 50)
print(len(articles))
df = reader.getArticlesDataframe()
else:
df = loadArticlesInEnglish()
df['text'] = ''
processor = Preprocessor(stopwordsList=None, lang=lang)
if (lang == 'russian'):
df['text'] = df['Source Text']
#for index, row in tqdm(df.iterrows(), total=df.shape[0]):
# df.at[index, 'text'] = processor.preprocess_text(row['Source Text'],
# useLemmas = False)
else:
df['text'] = df['Source Text']
subset = df.iloc[0:1, :]
for index, row in tqdm(subset.iterrows(), total=subset.shape[0]):
text_lemma_sw = processor.preprocess_text(row['Source Text'],
removeStopWords=True,
useLemmas=True)
text_lemma = processor.preprocess_text(row['Source Text'],
removeStopWords=False,
useLemmas=True)
subset.at[index, 'text_lemma_sw'] = text_lemma_sw
subset.at[
index,
'text_lemma'] = text_lemma #processor.preprocess_text(row['Source Text'], removeStopWords=False,
#useLemmas=True)
source = row['Source Text']
def object_generator(path, mass):
for s in os.listdir(path):
# find the .txt files
if '.txt' in s:
f_path = os.path.join(path, s)
yield rd.CV(filename=f_path, mass=mass)
def multi_cv(
ax,
path,
title=None,
mass=1.0,
color_set=['black', 'blue', 'green', 'red', 'pink', 'brown', 'yellow'],
xlabel=None,
ylabel=None,
filter_on=False,
smooth_range=None,
legend_font=12,
**kwargs):
'''Draw multiple cv curves on a single axe\
Arguments:\
ax\
path: the path of the file containing .txt files\
title\
mass: mass or surface area or volume of the sample'''
gp.Update_axe(ax, title=title, **kwargs)
# the list for Dataframe objects
cv = []
for s in os.listdir(path):
# find the .txt files
if '.txt' in s:
f_path = os.path.join(path, s)
obj = rd.CV(filename=f_path, mass=mass)
cv.append(obj)
# the list for line objects in an axe
for counter, obj in enumerate(cv):
# if filter is on
if filter_on == True and smooth_range is not None:
gp.MultiLine(smoother_cv(obj, smooth_range),
ax,
label=obj.scan_rate,
color=color_set[counter])
# or not...
else:
gp.MultiLine(obj,
ax,
label=obj.scan_rate,
color=color_set[counter])
if xlabel == None:
ax.set_xlabel(cv[0].columns[0])
else:
ax.set_xlabel(xlabel)
if ylabel == None:
ax.set_ylabel(cv[0].columns[1])
else:
ax.set_ylabel(ylabel)
gp.Auto_legend(ax,
bbox_to_anchor=(0, 1.03),
loc='upper left',
fontsize=legend_font)
def multi_c_d(
ax,
path,
title=None,
mass=1.0,
unit=' A/g',
color_set=['black', 'blue', 'green', 'red', 'pink', 'brown', 'yellow'],
xlabel=None,
ylabel=None,
legend_font=12,
b_to_a=(1, 0.75),
**kwargs):
'''Draw multiple charge/discharge curves on a single axe\
Arguments:\
ax\
path: the path of the file containing .txt files\
title\
mass: mass or surface area or volume of the sample'''
gp.Update_axe(ax, title=title, **kwargs)
# the list for GV objects
instances = []
for s in os.listdir(path):
# find the .txt files
if '.txt' in s:
f_path = os.path.join(path, s)
inst = rd.GV(filename=f_path, mass=mass)
instances.append(inst)
# the list for line objects in an axe
handles = []
for counter, inst in enumerate(instances):
l_char = gp.MultiLine(inst.charge_curve(),
ax,
label=str(inst.current) + unit,
color=color_set[counter])
gp.MultiLine(inst.discharge_curve(),
ax,
label=str(inst.current) + unit,
color=color_set[counter])
# line object is a list
handles += l_char
if xlabel == None:
ax.set_xlabel(instances[0].columns[0])
else:
ax.set_xlabel(xlabel)
if ylabel == None:
ax.set_ylabel(instances[0].columns[1])
else:
ax.set_ylabel(ylabel)
ax.set_xlim(left=0)
ax.set_ylim(top=0.8, bottom=0)
ax.legend(handles=handles,
bbox_to_anchor=b_to_a,
loc='upper right',
frameon=False,
fontsize=legend_font)
def test_ru_graph_keys_words_numbers():
lang = 'russian'
reader = Reader('elsevier journal.pdf') #, language='russian')
reader.loadFile()
words, lines, articles = reader.parseDocPages(
) #0, 50)#()#2160, 2190)#0, 50)
print(len(articles))
df_source = reader.getArticlesDataframe()
df = df_source.copy()
df['text'] = ''
df['noun_phrases_num'] = 0
processor = Preprocessor(stopwordsList=None, lang=lang)
sw = processor.stopwords
morph = pymorphy2.MorphAnalyzer()
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
text = processor.preprocess_text(row['Source Text'],
removeStopWords=True,
useLemmas=False)
df.at[index, 'text'] = text
pos = [morph.parse(w)[0].tag.POS for w in re.findall(r"[\w']+", text)]
count = Counter(pos)
df.at[
index,
'noun_phrases_num'] = count['NOUN'] + count['ADJF'] + count['PRTF']
df['keys_phrases_num'] = df.apply(
lambda row: len(row['Keywords'].split(',')), axis=1)
df['keys_words_num'] = df.apply(lambda row: len(
(re.findall(r"[\w']+", row['Keywords']))),
axis=1)
df['words_num_sw_incl'] = df.apply(lambda row: len(
(re.findall(r"[\w']+", row['Source Text']))),
axis=1)
df['words_num'] = df.apply(lambda row: len(
(re.findall(r"[\w']+", row['text']))),
axis=1)
stats = df[[
'keys_phrases_num', 'keys_words_num', 'words_num', 'words_num_sw_incl',
'noun_phrases_num'
]]
stats.to_excel("noun_phrases_num.xlsx")
#stats.to_excel("keys_words_number_stats.xlsx")
lst_phrases = stats['keys_phrases_num'].tolist()
lst_keys = stats['keys_words_num'].tolist()
lst_words = stats['words_num'].tolist()
lst_words_num_sw_incl = stats['words_num_sw_incl'].tolist()
lst_noun_phrases_num = stats['noun_phrases_num'].tolist()
x = list(range(1, len(lst_words) + 1))
plt.figure()
plt.plot(x, lst_phrases)
# Show/save figure as desired.
title = 'Количество ключевых фраз'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество ключевых фраз')
plt.savefig(title + " lst_phrases" + ".png", bbox_inches='tight')
plt.show()
plt.figure()
plt.plot(x, lst_keys)
# Show/save figure as desired.
title = 'Количество ключевых слов'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество ключевых слов')
plt.savefig(title + " lst_keys" + ".png", bbox_inches='tight')
plt.show()
plt.figure()
plt.plot(x, lst_words_num_sw_incl)
# Show/save figure as desired.
title = 'Количество слов в текстах'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество слов в тексте статьи')
plt.savefig(title + " words_num_sw_incl" + ".png", bbox_inches='tight')
plt.show()
plt.figure()
plt.plot(x, lst_words)
# Show/save figure as desired.
title = 'Количество слов в текстах (без стопслов)'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество слов в тексте статьи')
plt.savefig(title + " lst_words" + ".png", bbox_inches='tight')
plt.show()
plt.figure()
plt.plot(x, lst_noun_phrases_num)
# Show/save figure as desired.
title = 'Количество существительных, прилагательных и причастий в текстах'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество слов')
plt.savefig(title + " lst_noun_phrases_num" + ".png", bbox_inches='tight')
plt.show()
plt.figure()
plt.plot(x, lst_words_num_sw_incl, 'b', label='С учетом стопслов')
plt.plot(x, lst_words, 'g', label='Без учета стопслов')
plt.plot(x,
lst_noun_phrases_num,
'y',
label='Существительные, прилагательные и причастия')
# Show/save figure as desired.
title = 'Сравнение количества слов в текстах'
plt.title(title)
# Добавляем подписи к осям:
plt.xlabel("Номер статьи")
plt.ylabel('Количество слов')
plt.savefig(title + " compare_words_num_on_sw_pos" + ".png",
bbox_inches='tight')
plt.show()
def SimpleNN():
warnings.filterwarnings("ignore")
lang = 'russian'
nltk.download("stopwords")
lst_stopwords = stopwords.words("russian")
alphabet = "АаБбВвГгДдЕеЁёЖжЗзИиЙйКкЛлМмНнОоПпСсТтУуФфХхЦцЧчШшЩщЪъЫыЬьЭэЮюЯя"
alphabet += "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
lst_stopwords.extend(list(alphabet))
if lang == 'russian':
reader = Reader(
'elsevier journal.pdf',
generateWordList=True,
additional_stopwords=lst_stopwords) #, language='russian')
reader.loadFile()
words, lines, articles = reader.parseDocPages(4, 2190) #19, 400) #1)
df = pd.DataFrame.from_dict(words)
cat_features = [
'size',
'flags',
'font', #'color', #'otherSigns',
'morph_pos',
'morph_animacy',
'morph_aspect',
'morph_case',
'morph_gender',
'morph_involvement',
'morph_mood',
'morph_number',
'morph_person',
'morph_tense',
'morph_transitivity',
'morph_voice'
]
all_columns = list(df.columns)
df = pd.concat([
pd.get_dummies(df[col], prefix=col)
if col in cat_features and col != 'otherSigns' else df[col]
for col in all_columns
],
axis=1)
values = [',', '.', '\)', '\(', '\[', '\]']
for value in values:
df[str('otherSigns' + '_' + value)] = np.where(
df['otherSigns'].str.contains(value), "1", "0")
df = df.drop(['otherSigns'], axis=1)
print('all columns')
for col in all_columns:
print(col)
print()
print('new columns')
for col in df.columns:
print(col)
df.head()
featuresToRemove = [
'text',
'morph_normalform',
'morph_lexeme',
'span_count',
'line_count',
'block_count',
#'sentence_count',
'page_num',
'article_num',
'color'
]
#featuresToRemove = []
df = df.drop(featuresToRemove, axis=1)
# convert all columns of DataFrame
#df = df.apply(pd.to_numeric)
print("df.dtypes")
print(df.dtypes)
df.to_csv('all_words_features.csv', index=False)
numerics = ['int16', 'int32', 'int64', 'float16', 'float32', 'float64']
newdf = df.select_dtypes(include=numerics)
print("newdf.columns")
print(newdf.columns)
#df_y = newdf.pop('is_stopword')
keywords = newdf[newdf['is_keyword'] == 1]
print('len of keywords')
print(keywords.shape)
keywords.to_csv('all_keywords_features.csv', index=False)
try:
notkeywords = newdf[newdf['is_keyword'] == 0].sample(
n=len(keywords), replace=False)
except ValueError:
notkeywords = newdf[newdf['is_keyword'] == 0].sample(
n=len(keywords), replace=True)
print('len of notkeywords')
print(notkeywords.shape)
notkeywords.to_csv('all_notkeywords_features.csv', index=False)
newdf = pd.concat([notkeywords, keywords])
newdf = newdf.sample(frac=1)
df_Y = newdf.pop('is_keyword')
Y = df_Y.values
X = newdf.values.astype(float)
print('X.shape')
print(X.shape)
# evaluate model with standardized dataset
estimator = KerasClassifier(build_fn=create_baseline,
input_dim=len(newdf.columns),
epochs=100,
batch_size=5,
verbose=0)
kfold = StratifiedKFold(n_splits=10, shuffle=True)
results = cross_val_score(estimator, X, Y, cv=kfold, scoring='f1')
print("Baseline: %.2f%% (%.2f%%)" %
(results.mean() * 100, results.std() * 100))
#Baseline: 86.09% (4.30%)
#Baseline: 87.93% (1.14%)
#Baseline: 71.70% (3.28%)
#Baseline: 74.48% (1.10%) а1
print('bye-bye')
def main():
#lang = 'english'
lang = 'russian'
partMatchesCounted = False #if true then custom, else rough precision recall
num = 8
if lang == 'russian':
reader = Reader('elsevier journal.pdf') #, language='russian')
reader.loadFile()
#pages = reader.readPages();
##pages = reader.readPages(0, None, 'НАПРАВЛЕНИЕ 1', 'НАПРАВЛЕНИЕ 2',
##caseSensitiveSearch=True)
##pages = reader.readPages(0, None, 'НАПРАВЛЕНИЕ 1', 'НАПРАВЛЕНИЕ 2',
##caseSensitiveSearch=False)
##pages = reader.readPages(start=2, end=10, startString=None, endString=None,
##debug=True)
##pages = reader.readPages(start=4, end=4, startString=None, endString=None,
##debug=True)
#print(len(pages))
words, lines, articles = reader.parseDocPages() #2160, 2190)#0, 50)
print(len(articles))
df = reader.getArticlesDataframe()
else:
df = loadArticlesInEnglish()
df['text'] = ''
processor = Preprocessor(stopwordsList=None, lang=lang)
if (lang == 'russian'):
df['text'] = df['Source Text']
#for index, row in tqdm(df.iterrows(), total=df.shape[0]):
# df.at[index, 'text'] = processor.preprocess_text(row['Source Text'],
# useLemmas = False)
else:
df['text'] = df['Source Text']
#for index, row in tqdm(df.iterrows(), total=df.shape[0]):
#df.at[index, 'text'] = processor.preprocess_text(row['Source Text'],
#removeStopWords=False, useLemmas=False)
print(df.head())
rakeExtractor.extractKeywords(df,
num=num,
metricsCount=True,
partMatchesCounted=partMatchesCounted)
print(df.head())
#print(tabulate(df, headers='keys', tablefmt='psql'))
kw = df['Keywords'].values
#print(kw)
#print(df.Keywords)
tfidfBlobExtractor = TfIdfBlobExtractor(processor.getStopwords())
tfidfBlobExtractor.extractKeywords(df,
num=num,
metricsCount=True,
partMatchesCounted=partMatchesCounted)
#print(df.head())
textRankExtractor = TextRankExtractor(processor.getStopwords(),
language=lang)
textRankExtractor.extractKeywords(
df, num=num, metricsCount=True,
partMatchesCounted=partMatchesCounted) #False)
#print(df.head())
#print(tabulate(df.head(), headers='keys', tablefmt='psql'))
x = []
y = {
'rake': {
'precision': [],
'recall': [],
'f1': []
},
'textrank': {
'precision': [],
'recall': [],
'f1': []
},
'tfidf': {
'precision': [],
'recall': [],
'f1': []
}
}
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
x.append(index)
values = row['textrank_metrics'].split(',')
y['textrank']['precision'].append(values[0])
y['textrank']['recall'].append(values[1])
y['textrank']['f1'].append(values[2])
values = row['tfidf_blob_metrics'].split(',')
y['tfidf']['precision'].append(values[0])
y['tfidf']['recall'].append(values[1])
y['tfidf']['f1'].append(values[2])
values = row['rake_metrics'].split(',')
y['rake']['precision'].append(values[0])
y['rake']['recall'].append(values[1])
y['rake']['f1'].append(values[2])
#plt.plot(x, y['textrank']['precision'], 'g^', x, y['textrank']['recall'],
#'g-')
#fig, ax = plt.subplots()
#bar_values = [statistics.mean(list(map(float, y['textrank']['precision']))),
# statistics.mean(list(map(float, y['rake']['precision']))),
# statistics.mean(list(map(float, y['tfidf']['precision'])))]
#bar_label = bar_values
#bar_plot = plt.bar(['textrank', 'rake', 'tf-idf'], bar_values)
#autolabel(ax, bar_plot, bar_label)
#plt.ylim(0,max(bar_label) * 1.5)
#plt.title('Quality metrics for ' + lang + ' language')
#plt.savefig("add_text_bar_matplotlib_01.png", bbox_inches='tight')
#plt.show()
metrics = ['precision', 'recall', 'f1']
for i in range(len(metrics)):
fig, ax = plt.subplots()
bar_values = [
statistics.mean(list(map(float, y['textrank'][metrics[i]]))),
statistics.mean(list(map(float, y['rake'][metrics[i]]))),
statistics.mean(list(map(float, y['tfidf'][metrics[i]])))
]
bar_label = bar_values
bar_plot = plt.bar(['textrank', 'rake', 'tf-idf'], bar_values)
autolabel(ax, bar_plot, bar_label)
plt.ylim(0, max(max(bar_label) * 1.5, 0.01))
title = 'Metric ' + str(metrics[i] + ' for ' + str(num) +
' found keywords based on data in ' + lang +
' (partMatchesCounted = ' +
str(partMatchesCounted) + ')')
plt.title(title) #'Quality metrics for ' + lang + ' language')
plt.savefig(title + ".png", bbox_inches='tight')
plt.show()
def lemmaRuWithQuality():
lang = 'russian'
if lang == 'russian':
reader = Reader('elsevier journal.pdf') #, language='russian')
reader.loadFile()
words, lines, articles = reader.parseDocPages(
) #0, 50)#()#2160, 2190)#0, 50)
print(len(articles))
df_source = reader.getArticlesDataframe()
else:
df_source = loadArticlesInEnglish()
#partMatchesCounted = False #if true then custom, else rough precision recall
#num = 8;
#removeStopWords = False
conditions = list(
itertools.product(
[False, True], #[False, True], #removeStopWords
[
False, True
], #partMatchesCounted #if true then custom, else rough precision recall
[4, 8] #num of keywords
))
df = None
condition_i = 0
for condition in conditions:
removeStopWords = condition[0]
partMatchesCounted = condition[
1] #if true then custom, else rough precision recall
num = condition[2]
print('condition: ', condition)
if df is None:
print('Read DF')
df = df_source.copy()
df['text'] = ''
processor = Preprocessor(stopwordsList=None, lang=lang)
sw = processor.stopwords
#processor.stopwords = processor.get_normal_form_list(sw)
processor.stopwords.extend(processor.get_normal_form_list(sw))
if (lang == 'russian'):
#df['text'] = df['Source Text']
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
df.at[index, 'text'] = processor.preprocess_text(
row['Source Text'],
removeStopWords=removeStopWords,
useLemmas=True,
applyLemmasToText=True)
else:
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
text = processor.preprocess_text(
row['Source Text'],
removeStopWords=removeStopWords,
useLemmas=True)
df.at[index, 'text'] = text
elif condition_i == 4:
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
text = processor.preprocess_text(row['text'],
removeStopWords=True,
useLemmas=True,
applyLemmasToText=False)
df.at[index, 'text'] = text
condition_i = condition_i + 1
rakeExtractor.extractKeywords(df,
num=num,
metricsCount=True,
partMatchesCounted=partMatchesCounted,
textprocessor=processor)
tfidfBlobExtractor = TfIdfBlobExtractor(processor.getStopwords(),
textprocessor=processor)
tfidfBlobExtractor.extractKeywords(
df,
num=num,
metricsCount=True,
partMatchesCounted=partMatchesCounted)
textRankExtractor = TextRankExtractor(processor.getStopwords(),
language=lang,
textprocessor=processor)
textRankExtractor.extractKeywords(
df,
num=num,
metricsCount=True,
partMatchesCounted=partMatchesCounted)
x = []
y = {
'rake': {
'precision': [],
'recall': [],
'f1': []
},
'textrank': {
'precision': [],
'recall': [],
'f1': []
},
'tfidf': {
'precision': [],
'recall': [],
'f1': []
}
}
for index, row in tqdm(df.iterrows(), total=df.shape[0]):
x.append(index)
values = row['textrank_metrics'].split(',')
y['textrank']['precision'].append(values[0])
y['textrank']['recall'].append(values[1])
y['textrank']['f1'].append(values[2])
values = row['tfidf_blob_metrics'].split(',')
y['tfidf']['precision'].append(values[0])
y['tfidf']['recall'].append(values[1])
y['tfidf']['f1'].append(values[2])
values = row['rake_metrics'].split(',')
y['rake']['precision'].append(values[0])
y['rake']['recall'].append(values[1])
y['rake']['f1'].append(values[2])
metrics = ['precision', 'recall', 'f1']
for i in range(len(metrics)):
fig, ax = plt.subplots()
bar_values = [
statistics.mean(list(map(float, y['textrank'][metrics[i]]))),
statistics.mean(list(map(float, y['rake'][metrics[i]]))),
statistics.mean(list(map(float, y['tfidf'][metrics[i]])))
]
bar_label = [round(bv, 2) for bv in bar_values] #add round
bar_plot = plt.bar(['textrank', 'rake', 'tf-idf'], bar_values)
autolabel(ax, bar_plot, bar_label)
plt.ylim(0, max(max(bar_label) * 1.5, 0.01))
title = (
'Metric ' +
str(metrics[i] + ' for ' + str(num) +
' found keywords based on lemmatizied data in ' + lang +
' (partMatchesCounted = ' + str(partMatchesCounted) + ')' +
' (removeStopWords = ' + str(removeStopWords) + ')'))
plt.title(title) #'Quality metrics for ' + lang + ' language')
plt.savefig(title + ".png", bbox_inches='tight')
#if want to show only once upon a program
plt.show()