avr23-cds-translation2 / tabs /exploration_tab.py
Demosthene-OR's picture
Retour en arrirer sure requirements, et ses2seq
ec595bb
raw
history blame
No virus
19 kB
import streamlit as st
import os
import pandas as pd
import collections
from nltk.tokenize import word_tokenize
from nltk import download
from ast import literal_eval
from translate_app import tr
if st.session_state.Cloud == 0:
# import nltk
import contextlib
import re
from nltk.corpus import stopwords
import warnings
warnings.filterwarnings('ignore')
# from PIL import Image
# import time
# import random
title = "Exploration et Preprocessing"
sidebar_name = "Exploration et Preprocessing"
dataPath = st.session_state.DataPath
# Indiquer si l'on veut enlever les stop words. C'est un processus long
stopwords_to_do = True
# Indiquer si l'on veut lemmatiser les phrases, un fois les stop words enlevés. C'est un processus long (approximativement 8 minutes)
lemmatize_to_do = True
# Indiquer si l'on veut calculer le score Bleu pour tout le corpus. C'est un processus très long long (approximativement 10 minutes pour les 10 dictionnaires)
bleu_score_to_do = True
# Première ligne à charger
first_line = 0
# Nombre maximum de lignes à charger
max_lines = 140000
if ((first_line+max_lines)>137860):
max_lines = max(137860-first_line ,0)
# Nombre maximum de ligne à afficher pour les DataFrame
max_lines_to_display = 50
download('punkt')
if st.session_state.Cloud == 0:
download('averaged_perceptron_tagger')
with contextlib.redirect_stdout(open(os.devnull, "w")):
download('stopwords')
@st.cache_data
def load_data(path):
input_file = os.path.join(path)
with open(input_file, "r", encoding="utf-8") as f:
data = f.read()
# On convertit les majuscules en minulcule
data = data.lower()
data = data.split('\n')
return data[first_line:min(len(data),first_line+max_lines)]
@st.cache_data
def load_preprocessed_data(path,data_type):
input_file = os.path.join(path)
if data_type == 1:
return pd.read_csv(input_file, encoding="utf-8", index_col=0)
else:
with open(input_file, "r", encoding="utf-8") as f:
data = f.read()
data = data.split('\n')
if data_type==0:
data=data[:-1]
elif data_type == 2:
data=[eval(i) for i in data[:-1]]
elif data_type ==3:
data2 = []
for d in data[:-1]:
data2.append(literal_eval(d))
data=data2
return data
@st.cache_data
def load_all_preprocessed_data(lang):
txt =load_preprocessed_data(dataPath+'/preprocess_txt_'+lang,0)
txt_split = load_preprocessed_data(dataPath+'/preprocess_txt_split_'+lang,3)
txt_lem = load_preprocessed_data(dataPath+'/preprocess_txt_lem_'+lang,0)
txt_wo_stopword = load_preprocessed_data(dataPath+'/preprocess_txt_wo_stopword_'+lang,0)
df_count_word = pd.concat([load_preprocessed_data(dataPath+'/preprocess_df_count_word1_'+lang,1), load_preprocessed_data(dataPath+'/preprocess_df_count_word2_'+lang,1)])
return txt, txt_split, txt_lem, txt_wo_stopword, df_count_word
#Chargement des textes complet dans les 2 langues
full_txt_en = load_data(dataPath+'/small_vocab_en')
full_txt_fr = load_data(dataPath+'/small_vocab_fr')
# Chargement du résultat du préprocessing, si st.session_state.reCalcule == False
if not st.session_state.reCalcule:
full_txt_en, full_txt_split_en, full_txt_lem_en, full_txt_wo_stopword_en, full_df_count_word_en = load_all_preprocessed_data('en')
full_txt_fr, full_txt_split_fr, full_txt_lem_fr, full_txt_wo_stopword_fr, full_df_count_word_fr = load_all_preprocessed_data('fr')
else:
def remove_stopwords(text, lang):
stop_words = set(stopwords.words(lang))
# stop_words will contain set all english stopwords
filtered_sentence = []
for word in text.split():
if word not in stop_words:
filtered_sentence.append(word)
return " ".join(filtered_sentence)
def clean_undesirable_from_text(sentence, lang):
# Removing URLs
sentence = re.sub(r"https?://\S+|www\.\S+", "", sentence )
# Removing Punctuations (we keep the . character)
REPLACEMENTS = [("..", "."),
(",", ""),
(";", ""),
(":", ""),
("?", ""),
('"', ""),
("-", " "),
("it's", "it is"),
("isn't","is not"),
("'", " ")
]
for old, new in REPLACEMENTS:
sentence = sentence.replace(old, new)
# Removing Digits
sentence= re.sub(r'[0-9]','',sentence)
# Removing Additional Spaces
sentence = re.sub(' +', ' ', sentence)
return sentence
def clean_untranslated_sentence(data1, data2):
i=0
while i<len(data1):
if data1[i]==data2[i]:
data1.pop(i)
data2.pop(i)
else: i+=1
return data1,data2
import spacy
nlp_en = spacy.load('en_core_web_sm')
nlp_fr = spacy.load('fr_core_news_sm')
def lemmatize(sentence,lang):
# Create a Doc object
if lang=='en':
nlp=nlp_en
elif lang=='fr':
nlp=nlp_fr
else: return
doc = nlp(sentence)
# Create list of tokens from given string
tokens = []
for token in doc:
tokens.append(token)
lemmatized_sentence = " ".join([token.lemma_ for token in doc])
return lemmatized_sentence
def preprocess_txt (data, lang):
word_count = collections.Counter()
word_lem_count = collections.Counter()
word_wosw_count = collections.Counter()
corpus = []
data_split = []
sentence_length = []
data_split_wo_stopwords = []
data_length_wo_stopwords = []
data_lem = []
data_lem_length = []
txt_en_one_string= ". ".join([s for s in data])
txt_en_one_string = txt_en_one_string.replace('..', '.')
txt_en_one_string = " "+clean_undesirable_from_text(txt_en_one_string, 'lang')
data = txt_en_one_string.split('.')
if data[-1]=="":
data.pop(-1)
for i in range(len(data)): # On enleve les ' ' qui commencent et finissent les phrases
if data[i][0] == ' ':
data[i]=data[i][1:]
if data[i][-1] == ' ':
data[i]=data[i][:-1]
nb_phrases = len(data)
# Création d'un tableau de mots (sentence_split)
for i,sentence in enumerate(data):
sentence_split = word_tokenize(sentence)
word_count.update(sentence_split)
data_split.append(sentence_split)
sentence_length.append(len(sentence_split))
# La lemmatisation et le nettoyage des stopword va se faire en batch pour des raisons de vitesse
# (au lieu de le faire phrase par phrase)
# Ces 2 processus nécéssitent de connaitre la langue du corpus
if lang == 'en': l='english'
elif lang=='fr': l='french'
else: l="unknown"
if l!="unknown":
# Lemmatisation en 12 lots (On ne peut lemmatiser + de 1 M de caractères à la fois)
data_lemmatized=""
if lemmatize_to_do:
n_batch = 12
batch_size = round((nb_phrases/ n_batch)+0.5)
for i in range(n_batch):
to_lem = ".".join([s for s in data[i*batch_size:(i+1)*batch_size]])
data_lemmatized = data_lemmatized+"."+lemmatize(to_lem,lang).lower()
data_lem_for_sw = data_lemmatized[1:]
data_lemmatized = data_lem_for_sw.split('.')
for i in range(nb_phrases):
data_lem.append(data_lemmatized[i].split())
data_lem_length.append(len(data_lemmatized[i].split()))
word_lem_count.update(data_lem[-1])
# Elimination des StopWords en un lot
# On élimine les Stopwords des phrases lémmatisés, si cette phase a eu lieu
# (wosw signifie "WithOut Stop Words")
if stopwords_to_do:
if lemmatize_to_do:
data_wosw = remove_stopwords(data_lem_for_sw,l)
else:
data_wosw = remove_stopwords(txt_en_one_string,l)
data_wosw = data_wosw.split('.')
for i in range(nb_phrases):
data_split_wo_stopwords.append(data_wosw[i].split())
data_length_wo_stopwords.append(len(data_wosw[i].split()))
word_wosw_count.update(data_split_wo_stopwords[-1])
corpus = list(word_count.keys())
# Création d'un DataFrame txt_n_unique_val :
# colonnes = mots
# lignes = phases
# valeur de la cellule = nombre d'occurence du mot dans la phrase
## BOW
from sklearn.feature_extraction.text import CountVectorizer
count_vectorizer = CountVectorizer(analyzer="word", ngram_range=(1, 1), token_pattern=r"[^' ']+" )
# Calcul du nombre d'apparition de chaque mot dans la phrases
countvectors = count_vectorizer.fit_transform(data)
corpus = count_vectorizer.get_feature_names_out()
txt_n_unique_val= pd.DataFrame(columns=corpus,index=range(nb_phrases), data=countvectors.todense()).astype(float)
return data, corpus, data_split, data_lemmatized, data_wosw, txt_n_unique_val, sentence_length, data_length_wo_stopwords, data_lem_length
def count_world(data):
word_count = collections.Counter()
for sentence in data:
word_count.update(word_tokenize(sentence))
corpus = list(word_count.keys())
nb_mots = sum(word_count.values())
nb_mots_uniques = len(corpus)
return corpus, nb_mots, nb_mots_uniques
def display_preprocess_results(lang, data, data_split, data_lem, data_wosw, txt_n_unique_val):
global max_lines, first_line, last_line, lemmatize_to_do, stopwords_to_do
corpus = []
nb_phrases = len(data)
corpus, nb_mots, nb_mots_uniques = count_world(data)
mots_lem, _ , nb_mots_lem = count_world(data_lem)
mots_wo_sw, _ , nb_mots_wo_stopword = count_world(data_wosw)
# Identifiez les colonnes contenant uniquement des zéros et les supprimer
columns_with_only_zeros = txt_n_unique_val.columns[txt_n_unique_val.eq(0).all()]
txt_n_unique_val = txt_n_unique_val.drop(columns=columns_with_only_zeros)
# Affichage du nombre de mot en fonction du pré-processing réalisé
tab1, tab2, tab3, tab4 = st.tabs([tr("Résumé"), tr("Tokenisation"),tr("Lemmatisation"), tr("Sans Stopword")])
with tab1:
st.subheader(tr("Résumé du pré-processing"))
st.write("**"+tr("Nombre de phrases")+" : "+str(nb_phrases)+"**")
st.write("**"+tr("Nombre de mots")+" : "+str(nb_mots)+"**")
st.write("**"+tr("Nombre de mots uniques")+" : "+str(nb_mots_uniques)+"**")
st.write("")
st.write("\n**"+tr("Nombre d'apparitions de chaque mot dans chaque phrase (:red[Bag Of Words]):")+"**")
st.dataframe(txt_n_unique_val.head(max_lines_to_display), width=800)
with tab2:
st.subheader(tr("Tokenisation"))
st.write(tr('Texte "splited":'))
st.dataframe(pd.DataFrame(data=data_split, index=range(first_line,last_line)).head(max_lines_to_display).fillna(''), width=800)
st.write("**"+tr("Nombre de mots uniques")+" : "+str(nb_mots_uniques)+"**")
st.write("")
st.write("\n**"+tr("Mots uniques")+":**")
st.markdown(corpus[:500])
st.write("\n**"+tr("Nombre d'apparitions de chaque mot dans chaque phrase (:red[Bag Of Words]):")+"**")
st.dataframe(txt_n_unique_val.head(max_lines_to_display), width=800)
with tab3:
st.subheader(tr("Lemmatisation"))
if lemmatize_to_do:
st.dataframe(pd.DataFrame(data=data_lem,columns=[tr('Texte lemmatisé')],index=range(first_line,last_line)).head(max_lines_to_display), width=800)
# Si langue anglaise, affichage du taggage des mots
# if lang == 'en':
# for i in range(min(5,len(data))):
# s = str(nltk.pos_tag(data_split[i]))
# st.markdown("**Texte avec Tags "+str(i)+"** : "+s)
st.write("**"+tr("Nombre de mots uniques lemmatisés")+" : "+str(nb_mots_lem)+"**")
st.write("")
st.write("\n**"+tr("Mots uniques lemmatisés:")+"**")
st.markdown(mots_lem[:500])
with tab4:
st.subheader(tr("Sans Stopword"))
if stopwords_to_do:
st.dataframe(pd.DataFrame(data=data_wosw,columns=['Texte sans stopwords'],index=range(first_line,last_line)).head(max_lines_to_display), width=800)
st.write("**"+tr("Nombre de mots uniques sans stop words")+": "+str(nb_mots_wo_stopword)+"**")
st.write("")
st.write("\n**"+tr("Mots uniques sans stop words")+":**")
st.markdown(mots_wo_sw[:500])
def run():
global max_lines, first_line, last_line, lemmatize_to_do, stopwords_to_do
global full_txt_en, full_txt_split_en, full_txt_lem_en, full_txt_wo_stopword_en, full_df_count_word_en
global full_txt_fr, full_txt_split_fr, full_txt_lem_fr, full_txt_wo_stopword_fr, full_df_count_word_fr
st.write("")
st.title(tr(title))
st.write("## **"+tr("Explications")+" :**\n")
st.markdown(tr(
"""
Le traitement du langage naturel permet à l'ordinateur de comprendre et de traiter les langues humaines.
Lors de notre projet, nous avons étudié le dataset small_vocab, proposés par Suzan Li, Chief Data Scientist chez Campaign Research à Toronto.
Celui-ci représente un corpus de phrases simples en anglais, et sa traduction (approximative) en français.
:red[**Small_vocab**] contient 137 860 phrases en anglais et français.
""")
, unsafe_allow_html=True)
st.markdown(tr(
"""
Afin de découvrir ce corpus et de préparer la traduction, nous allons effectuer un certain nombre de tâches de pré-traitement (preprocessing).
Ces taches sont, par exemple:
""")
, unsafe_allow_html=True)
st.markdown(
"* "+tr("le :red[**nettoyage**] du texte (enlever les majuscules et la ponctuation)")+"\n"+ \
"* "+tr("la :red[**tokenisation**] (découpage du texte en mots)")+"\n"+ \
"* "+tr("la :red[**lemmatisation**] (traitement lexical qui permet de donner une forme unique à toutes les \"variations\" d'un même mot)")+"\n"+ \
"* "+tr("l'élimination des :red[**mots \"transparents\"**] (sans utilité pour la compréhension, tels que les articles).")+" \n"+ \
tr("Ce prétraintement se conclut avec la contruction d'un :red[**Bag Of Worlds**], c'est à dire une matrice qui compte le nombre d'apparition de chaque mots (colonne) dans chaque phrase (ligne)")
, unsafe_allow_html=True)
#
st.write("## **"+tr("Paramètres")+" :**\n")
Langue = st.radio(tr('Langue:'),('Anglais','Français'), horizontal=True)
first_line = st.slider(tr('No de la premiere ligne à analyser:'),0,137859)
max_lines = st.select_slider(tr('Nombre de lignes à analyser:'),
options=[1,5,10,15,100, 500, 1000,'Max'])
if max_lines=='Max':
max_lines=137860
if ((first_line+max_lines)>137860):
max_lines = max(137860-first_line,0)
last_line = first_line+max_lines
if (Langue=='Anglais'):
st.dataframe(pd.DataFrame(data=full_txt_en,columns=['Texte']).loc[first_line:last_line-1].head(max_lines_to_display), width=800)
else:
st.dataframe(pd.DataFrame(data=full_txt_fr,columns=['Texte']).loc[first_line:last_line-1].head(max_lines_to_display), width=800)
st.write("")
# Chargement des textes sélectionnés dans les 2 langues (max lignes = max_lines)
txt_en = full_txt_en[first_line:last_line]
txt_fr = full_txt_fr[first_line:last_line]
# Elimination des phrases non traduites
# txt_en, txt_fr = clean_untranslated_sentence(txt_en, txt_fr)
if not st.session_state.reCalcule:
txt_split_en = full_txt_split_en[first_line:last_line]
txt_lem_en = full_txt_lem_en[first_line:last_line]
txt_wo_stopword_en = full_txt_wo_stopword_en[first_line:last_line]
df_count_word_en = full_df_count_word_en.loc[first_line:last_line-1]
txt_split_fr = full_txt_split_fr[first_line:last_line]
txt_lem_fr = full_txt_lem_fr[first_line:last_line]
txt_wo_stopword_fr = full_txt_wo_stopword_fr[first_line:last_line]
df_count_word_fr = full_df_count_word_fr.loc[first_line:last_line-1]
# Lancement du préprocessing du texte qui va spliter nettoyer les phrases et les spliter en mots
# et calculer nombre d'occurences des mots dans chaque phrase
if (Langue == 'Anglais'):
st.write("## **"+tr("Préprocessing de small_vocab_en")+" :**\n")
if max_lines>10000:
with st.status(":sunglasses:", expanded=True):
if st.session_state.reCalcule:
txt_en, corpus_en, txt_split_en, txt_lem_en, txt_wo_stopword_en, df_count_word_en,sent_len_en, sent_wo_sw_len_en, sent_lem_len_en = preprocess_txt (txt_en,'en')
display_preprocess_results('en',txt_en, txt_split_en, txt_lem_en, txt_wo_stopword_en, df_count_word_en)
else:
if st.session_state.reCalcule:
txt_en, corpus_en, txt_split_en, txt_lem_en, txt_wo_stopword_en, df_count_word_en,sent_len_en, sent_wo_sw_len_en, sent_lem_len_en = preprocess_txt (txt_en,'en')
display_preprocess_results('en',txt_en, txt_split_en, txt_lem_en, txt_wo_stopword_en, df_count_word_en)
else:
st.write("## **"+tr("Préprocessing de small_vocab_fr")+" :**\n")
if max_lines>10000:
with st.status(":sunglasses:", expanded=True):
if st.session_state.reCalcule:
txt_fr, corpus_fr, txt_split_fr, txt_lem_fr, txt_wo_stopword_fr, df_count_word_fr,sent_len_fr, sent_wo_sw_len_fr, sent_lem_len_fr = preprocess_txt (txt_fr,'fr')
display_preprocess_results('fr', txt_fr, txt_split_fr, txt_lem_fr, txt_wo_stopword_fr, df_count_word_fr)
else:
if st.session_state.reCalcule:
txt_fr, corpus_fr, txt_split_fr, txt_lem_fr, txt_wo_stopword_fr, df_count_word_fr,sent_len_fr, sent_wo_sw_len_fr, sent_lem_len_fr = preprocess_txt (txt_fr,'fr')
display_preprocess_results('fr', txt_fr, txt_split_fr, txt_lem_fr, txt_wo_stopword_fr, df_count_word_fr)