Deep learning model, heat map, data prepo

Related tags

Deep LearningDemocrate_debat_usa
Overview

DEEP LEARNING ON USA DEMOCRATES DEBATE

By Pamela Dekas

import sys
import csv
import re 
import nltk
import string
import unicodedata
from textblob import TextBlob
from collections import Counter
import pandas as pd
import numpy as np
from wordcloud import WordCloud
from nltk.classify import * 
from nltk.corpus import stopwords
from sklearn.metrics import f1_score, roc_auc_score
from sklearn.feature_extraction.text import CountVectorizer
from nltk.tokenize import word_tokenize
import nltk.classify.util
import matplotlib.pyplot as plt
from string import punctuation 
from nltk.corpus import stopwords
from wordcloud import STOPWORDS
import os
from sklearn.model_selection import train_test_split
from keras.datasets import imdb
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM
from keras.layers.embeddings import Embedding
from keras.preprocessing import sequence, text
from keras.callbacks import EarlyStopping
Using TensorFlow backend.



---------------------------------------------------------------------------

AttributeError                            Traceback (most recent call last)


   
     in 
    
     ()
     22 import os
     23 from sklearn.model_selection import train_test_split
---> 24 from keras.datasets import imdb
     25 from keras.models import Sequential
     26 from keras.layers import Dense


~\Anaconda3\lib\site-packages\keras\__init__.py in 
     
      ()
      1 from __future__ import absolute_import
      2 
----> 3 from . import utils
      4 from . import activations
      5 from . import applications


~\Anaconda3\lib\site-packages\keras\utils\__init__.py in 
      
       ()
      4 from . import data_utils
      5 from . import io_utils
----> 6 from . import conv_utils
      7 from . import losses_utils
      8 from . import metrics_utils


~\Anaconda3\lib\site-packages\keras\utils\conv_utils.py in 
       
        () 7 from six.moves import range 8 import numpy as np ----> 9 from .. import backend as K 10 11 ~\Anaconda3\lib\site-packages\keras\backend\__init__.py in 
        
         () ----> 1 from .load_backend import epsilon 2 from .load_backend import set_epsilon 3 from .load_backend import floatx 4 from .load_backend import set_floatx 5 from .load_backend import cast_to_floatx ~\Anaconda3\lib\site-packages\keras\backend\load_backend.py in 
         
          () 88 elif _BACKEND == 'tensorflow': 89 sys.stderr.write('Using TensorFlow backend.\n') ---> 90 from .tensorflow_backend import * 91 else: 92 # Try and load external backend. ~\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py in 
          
           () 52 53 # Private TF Keras utils ---> 54 get_graph = tf_keras_backend.get_graph 55 # learning_phase_scope = tf_keras_backend.learning_phase_scope # TODO 56 name_scope = tf.name_scope AttributeError: module 'tensorflow.python.keras.backend' has no attribute 'get_graph' 
          
         
        
       
      
     
    
   
speech = pd.read_csv('debate_transcripts_v3_2020-02-26.csv',encoding= 'unicode_escape')
df= pd.DataFrame(speech)
dem_speakers = df["speaker"]
number_of_speakers = len(set(dem_speakers))
print("Nombre de speakers:",number_of_speakers, "speakers")

# Mean duration of speech.
print("temps moyen de parole:",np.mean(df["speaking_time_seconds"]), "seconds")
print("Dataset size:", len(df))
Nombre de speakers: 106 speakers
temps moyen de parole: 16.49230769230769 seconds
Dataset size: 5911

df.info()

   
    
RangeIndex: 5911 entries, 0 to 5910
Data columns (total 6 columns):
date                     5911 non-null object
debate_name              5911 non-null object
debate_section           5911 non-null object
speaker                  5911 non-null object
speech                   5911 non-null object
speaking_time_seconds    5395 non-null float64
dtypes: float64(1), object(5)
memory usage: 277.2+ KB

   
df.groupby('speaker')['speaking_time_seconds'].sum(level=0).nlargest(10).plot.bar()
plt.title('Repartition par temps de parole')
plt.show()

png

debate_time = df.groupby(by=['speaker', 'date']).speaking_time_seconds.sum().nlargest(15)
debate_time.plot()

png

suppresion des colonnes qui ne seront pas utilisé dans la suite du projet et creation du dataset final###

df=df.drop(['date','debate_name','debate_section','speaking_time_seconds'],1)
df.head(5)
speaker speech
0 Norah O�Donnell Good evening and welcome, the Democratic presi...
1 Gayle King And Super Tuesday is just a week away and this...
2 Norah O�Donnell And CBS News is proud to bring you this debate...
3 Gayle King And we are partnering tonight also with Twitte...
4 Norah O�Donnell Now, here are the rules for the next two hours...

PREPROCESSING 

import nltk 
nltk.download('punkt')
stopwords = nltk.corpus.stopwords.words('english')
Tailored_stopwords=('im','ive','mr','weve','dont','well','will','make','us','we',
                      'I','make','got','need','want','think',
                      'going','go','one','thank','going',
                      'way','say','every','re','us','first',
                     'now','said','know','look','done','take',
                     'number','two','three','s','m',"t",
                      'let','don','tell','ve','im','mr','put','maybe','whether','many', 'll','around','thing','Secondly','doesn','lot')
#stopwords = nltk.corpus.stopwords.words('english')
stopwords = set(STOPWORDS)
stopwords= stopwords.union(Tailored_stopwords)
[nltk_data] Downloading package punkt to C:\Users\pamel.DESKTOP-O19M7N
[nltk_data]     F\AppData\Roaming\nltk_data...
[nltk_data]   Package punkt is already up-to-date!

def Text_cleansing(speech):
    speech = re.sub('@[A-Za-z0–9]+', '', str(speech))
    speech = re.sub('#', '', speech) # Enlever les '#' hash tag
    speech = re.sub('rt', '', speech)
    speech=re.sub(',',' ', speech)
    speech=re.sub('!',' ',speech)
    speech=re.sub(':',' ',speech)
    speech=re.sub("'","",speech)
    speech=re.sub('"','',speech)
    speech=speech.lower()
    speech = word_tokenize(speech)
    return speech
def remove_stopwords(speech):
    speech_clean = [word for word in speech if word not in stopwords]
    return speech_clean
                         
df['speech_tokens']= df['speech'].apply(Text_cleansing)
df.head(5)
speaker speech speech_tokens
0 Norah O�Donnell Good evening and welcome, the Democratic presi... [good, evening, and, welcome, the, democratic,...
1 Gayle King And Super Tuesday is just a week away and this... [and, super, tuesday, is, just, a, week, away,...
2 Norah O�Donnell And CBS News is proud to bring you this debate... [and, cbs, news, is, proud, to, bring, you, th...
3 Gayle King And we are partnering tonight also with Twitte... [and, we, are, panering, tonight, also, with, ...
4 Norah O�Donnell Now, here are the rules for the next two hours... [now, here, are, the, rules, for, the, next, t... 
df['speech_clean']=df['speech_tokens'].apply(remove_stopwords)
df.head(5)
speaker speech speech_tokens speech_clean
0 Norah O�Donnell Good evening and welcome, the Democratic presi... [good, evening, and, welcome, the, democratic,... [good, evening, welcome, democratic, president...
1 Gayle King And Super Tuesday is just a week away and this... [and, super, tuesday, is, just, a, week, away,... [super, tuesday, week, away, biggest, primary,...
2 Norah O�Donnell And CBS News is proud to bring you this debate... [and, cbs, news, is, proud, to, bring, you, th... [cbs, news, proud, bring, debate, along, co-sp...
3 Gayle King And we are partnering tonight also with Twitte... [and, we, are, panering, tonight, also, with, ... [panering, tonight, twitter, ., home, paicipat...
4 Norah O�Donnell Now, here are the rules for the next two hours... [now, here, are, the, rules, for, the, next, t... [rules, next, hours, ., asked, question, minut... 
def wordcloud(dataframe):
    Aw= df['speech_clean']
    wordCloud = WordCloud(width=500, height=300,background_color='white', max_font_size=110).generate(str(Aw))
    plt.imshow(wordCloud, interpolation="bilinear")
    plt.axis("off")
    plt.title("speech wordcloud")

wordcloud(df['speech_clean'])

png

Pour la suite du projet on reduira la liste des speakers aux candidats les plus notoires (top 7 speakers)###

df = df.loc[df.speaker.isin({'Joe Biden', 'Bernie Sanders', 'Elizabeth Warren', 'Michael Bloomberg', 'Pete Buttigieg', 'Amy Klobuchar',  'Tulsi Gabbard'})]
df.head()
df.shape
(2245, 4)

CountVectorizer et creation du dict des mots par candidat a utiliser sur les modeles ML qui seront en back-up###

Analyse Lexicale

cv = CountVectorizer(stop_words=stopwords)
df_cv = cv.fit_transform(df.speech)
df_words = pd.DataFrame(df_cv.toarray(), columns=cv.get_feature_names())
df_words.index = df.speaker
df_words = df_words.transpose()
df_words
speaker Bernie Sanders Michael Bloomberg Michael Bloomberg Bernie Sanders Pete Buttigieg Elizabeth Warren Elizabeth Warren Pete Buttigieg Joe Biden Bernie Sanders ... Amy Klobuchar Elizabeth Warren Amy Klobuchar Tulsi Gabbard Tulsi Gabbard Amy Klobuchar Amy Klobuchar Amy Klobuchar Elizabeth Warren Elizabeth Warren
00 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
000 2 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
001st 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
01 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
02 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
03 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
04 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
05 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
06 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
07 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
08 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
09 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
100 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
10000 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
100s 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
10th 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 1 0 ... 0 0 0 0 0 0 0 0 0 0
120 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
125 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
12th 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
130 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
135 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
137 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
13th 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
140 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
149 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
xinjiang 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yachts 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yale 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yang 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yanked 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
ye 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yeah 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 1 0 0
year 1 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yearly 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
years 2 0 0 0 0 0 0 1 0 0 ... 0 0 0 0 0 0 0 0 1 0
yellow 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yemen 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yemin 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yep 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yes 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yesterday 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yet 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yo 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
york 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
yorker 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
young 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
younger 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
youngest 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
youth 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
youtube 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
zealand 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
zero 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
zeroed 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
zip 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0
zone 0 0 0 0 0 0 0 0 0 0 ... 0 0 0 0 0 0 0 0 0 0

6385 rows × 2245 columns

top_dict = {}
for c in df_words.columns:
    top = df_words[c].sort_values(ascending=False).head(30)
    top_dict[c]= list(zip(top.index, top.values))
for speaker, top_words in top_dict.items():
    print(speaker)
    print(', '.join([word for word, count in top_words[0:9]]))
    print('---')
---------------------------------------------------------------------------

TypeError                                 Traceback (most recent call last)


   
     in 
    
     ()
      1 top_dict = {}
      2 for c in df_words.columns:
----> 3     top = df_words[c].sort_values(ascending=False).head(30)
      4     top_dict[c]= list(zip(top.index, top.values))
      5 for speaker, top_words in top_dict.items():


TypeError: sort_values() missing 1 required positional argument: 'by'

    
   
df2=pd.DataFrame(top_dict)
df2.head(15)
 
from collections import Counter
words = []
for speaker in df_words.columns:
    top = [word for (word, count) in top_dict[speaker]]
    for t in top:
        words.append(t)
Counter(words).most_common(15)
---------------------------------------------------------------------------

KeyError                                  Traceback (most recent call last)


   
     in 
    
     ()
      2 words = []
      3 for speaker in df_words.columns:
----> 4     top = [word for (word, count) in top_dict[speaker]]
      5     for t in top:
      6         words.append(t)


KeyError: 'Bernie Sanders'

Implemantation du modèle###

print(df.columns)
print(df.shape)
df['speaker'] = df['speaker'].astype(str)
Index(['speaker', 'speech', 'speech_tokens', 'speech_clean'], dtype='object')
(2245, 4)

Embedding

import gensim
RANDOM_STATE = 50
EPOCHS = 5
BATCH_SIZE = 256
EMB_DIM = 100
SAVE_MODEL = True

X = df['speech_clean']
print(X.head())
X.shape
5     [well, you�re, right, economy, really, great, ...
6                                            [senator-]
8     [think, donald, trump, thinks, would, better, ...
9     [oh, mr., bloomberg, ., let, tell, mr., putin,...
11     [know, president, russia, wants, it�s, chaos, .]
Name: speech_clean, dtype: object





(2245,)

emb_model = gensim.models.Word2Vec(sentences = X, size = EMB_DIM, window = 5, workers = 4, min_count = 1)
print('La taille du vocabulaire appris est de ',len(list(emb_model.wv.vocab)))
La taille du vocabulaire appris est de  7139

from keras.preprocessing.text import Tokenizer
import tokenize
max_length = max([len(s) for s in X])

tokenizer_new = Tokenizer()
tokenizer_new.fit_on_texts(X)

X_seq = tokenizer_new.texts_to_sequences(X)
X_fin = sequence.pad_sequences(X_seq, maxlen = max_length)
print(X_fin.shape)
(2245, 140)

emb_vec = emb_model.wv
MAX_NB_WORDS = len(list(emb_vec.vocab))
tokenizer_word_index = tokenizer_new.word_index
vocab_size = len(tokenizer_new.word_index) + 1
embedded_matrix = np.zeros((vocab_size, EMB_DIM))


for word, i in tokenizer_word_index.items():
    if i>= MAX_NB_WORDS:
        continue
    try:
        embedding_vector = emb_vec[word]
        wv_matrix[i] = embedding_vector
    except:
        pass      
embedded_matrix.shape
print(embedded_matrix)
[[0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 ...
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]]

Préparation des variables

from keras.utils import to_categorical
from sklearn.preprocessing import LabelEncoder
y = df.speaker
print(y.head(10))
y.shape
5     1
6     4
8     4
9     1
11    5
12    2
13    2
15    5
21    3
23    1
Name: speaker, dtype: int32





(2245,)

Counter(y)
Counter({'Bernie Sanders': 430,
         'Michael Bloomberg': 97,
         'Pete Buttigieg': 392,
         'Elizabeth Warren': 440,
         'Joe Biden': 456,
         'Amy Klobuchar': 353,
         'Tulsi Gabbard': 77})

le=LabelEncoder()
df['speaker'] = le.fit_transform(df['speaker'])
df.head()

y = df.speaker
y.head()
print(y.shape)
print(X_fin.shape)
(2245,)
(2245, 140)

X_train, X_test, y_train, y_test = train_test_split(X_fin , y, test_size = 0.2, random_state = 42)


print(X_train.shape)
print(y_train.shape)
(1796, 140)
(1796,)

Construction des NN

model_pre_trained = Sequential()

model_pre_trained.add(Embedding(vocab_size, EMB_DIM, weights = [embedded_matrix], 
                    input_length = max_length, trainable = False))
model_pre_trained.add(LSTM(128, dropout=0.2, recurrent_dropout=0.2))
model_pre_trained.add(Dense(1, activation='softmax'))

model_pre_trained.compile(loss='categorical_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])

model_pre_trained.summary()
Model: "sequential_11"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
embedding_11 (Embedding)     (None, 140, 100)          714000    
_________________________________________________________________
lstm_13 (LSTM)               (None, 128)               117248    
_________________________________________________________________
dense_9 (Dense)              (None, 1)                 129       
=================================================================
Total params: 831,377
Trainable params: 117,377
Non-trainable params: 714,000
_________________________________________________________________

Fitting

history_pre_trained = model_pre_trained.fit(X_fin, y, batch_size = BATCH_SIZE, epochs =20, verbose =1, validation_split = 0.2)
Train on 1796 samples, validate on 449 samples
Epoch 1/20
1796/1796 [==============================] - 4s 2ms/step - loss: 0.5429 - accuracy: 0.1754 - val_loss: -0.4417 - val_accuracy: 0.2472
Epoch 2/20
1796/1796 [==============================] - 3s 2ms/step - loss: -6.7429 - accuracy: 0.1776 - val_loss: -14.1017 - val_accuracy: 0.2472
Epoch 3/20
1796/1796 [==============================] - 3s 2ms/step - loss: -15.8550 - accuracy: 0.1776 - val_loss: -19.5441 - val_accuracy: 0.2472
Epoch 4/20
1796/1796 [==============================] - 3s 2ms/step - loss: -20.7949 - accuracy: 0.1776 - val_loss: -23.4335 - val_accuracy: 0.2472
Epoch 5/20
1796/1796 [==============================] - 3s 2ms/step - loss: -24.1430 - accuracy: 0.1776 - val_loss: -25.9735 - val_accuracy: 0.2472
Epoch 6/20
1796/1796 [==============================] - 3s 2ms/step - loss: -26.4535 - accuracy: 0.1776 - val_loss: -28.0725 - val_accuracy: 0.2472
Epoch 7/20
1796/1796 [==============================] - 3s 2ms/step - loss: -28.4266 - accuracy: 0.1776 - val_loss: -29.9313 - val_accuracy: 0.2472
Epoch 8/20
1796/1796 [==============================] - 3s 2ms/step - loss: -30.1754 - accuracy: 0.1776 - val_loss: -31.6261 - val_accuracy: 0.2472
Epoch 9/20
1796/1796 [==============================] - 3s 2ms/step - loss: -31.8791 - accuracy: 0.1776 - val_loss: -33.3337 - val_accuracy: 0.2472
Epoch 10/20
1796/1796 [==============================] - 4s 2ms/step - loss: -33.5166 - accuracy: 0.1776 - val_loss: -34.9834 - val_accuracy: 0.2472
Epoch 11/20
1796/1796 [==============================] - 3s 2ms/step - loss: -35.1544 - accuracy: 0.1776 - val_loss: -36.5973 - val_accuracy: 0.2472
Epoch 12/20
1796/1796 [==============================] - 3s 2ms/step - loss: -36.7253 - accuracy: 0.1776 - val_loss: -38.2070 - val_accuracy: 0.2472
Epoch 13/20
1796/1796 [==============================] - 3s 2ms/step - loss: -38.3344 - accuracy: 0.1776 - val_loss: -39.8655 - val_accuracy: 0.2472
Epoch 14/20
1796/1796 [==============================] - 3s 2ms/step - loss: -39.9810 - accuracy: 0.1776 - val_loss: -41.5162 - val_accuracy: 0.2472
Epoch 15/20
1796/1796 [==============================] - 3s 1ms/step - loss: -41.6567 - accuracy: 0.1776 - val_loss: -43.2049 - val_accuracy: 0.2472
Epoch 16/20
1796/1796 [==============================] - 3s 1ms/step - loss: -43.2579 - accuracy: 0.1776 - val_loss: -44.8235 - val_accuracy: 0.2472
Epoch 17/20
1796/1796 [==============================] - 3s 1ms/step - loss: -44.9030 - accuracy: 0.1776 - val_loss: -46.4982 - val_accuracy: 0.2472
Epoch 18/20
1796/1796 [==============================] - 2s 1ms/step - loss: -46.5038 - accuracy: 0.1776 - val_loss: -48.0627 - val_accuracy: 0.2472
Epoch 19/20
1796/1796 [==============================] - 3s 1ms/step - loss: -48.0124 - accuracy: 0.1776 - val_loss: -49.5424 - val_accuracy: 0.2472
Epoch 20/20
1796/1796 [==============================] - 2s 1ms/step - loss: -49.5209 - accuracy: 0.1776 - val_loss: -51.1489 - val_accuracy: 0.2472

Evaluation du modèle

score = model_pre_trained.evaluate(X_test, y_test, verbose = 0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
Test loss: -51.148848297866785
Test accuracy: 0.18930958211421967

ptoblèmes: npmbre important de stopwords à rajouter au dictionnaire, doutes sur la fonction dactivation, stemming/lemmatization qui semble peu efficace; axes d'amélioration: explorer les N grammes pouir contextualiser les mots et creer u_n dictionnaire de stopwords customisé pour les deabts ( association d'idées)/






























Owner









Pamela Dekas



Adepte de text mining, deep learning and data visualization



Pamela Dekas




 GitHub Repository










Lux AI environment interface for RLlib multi-agents


 Lux AI interface to RLlib MultiAgentsEnv For Lux AI Season 1 Kaggle competition. LuxAI repo RLlib-multiagents docs Kaggle environments repo Please let





Jaime
 12  Nov 07, 2022









Clean and readable code for Decision Transformer: Reinforcement Learning via Sequence Modeling


 Minimal implementation of Decision Transformer: Reinforcement Learning via Sequence Modeling in PyTorch for mujoco control tasks in OpenAI gym





Nikhil Barhate
 104  Jan 06, 2023









Ensemble Knowledge Guided Sub-network Search and Fine-tuning for Filter Pruning


 Ensemble Knowledge Guided Sub-network Search and Fine-tuning for Filter Pruning This repository is official Tensorflow implementation of paper: Ensemb





Seunghyun Lee
 12  Oct 18, 2022









Connecting Java/ImgLib2 + Python/NumPy


 imglyb imglyb aims at connecting two worlds that have been seperated for too long: Python with numpy Java with ImgLib2 imglyb uses jpype to access num





ImgLib2
 29  Dec 21, 2022









Official code for "Decoupling Zero-Shot Semantic Segmentation"


 Decoupling Zero-Shot Semantic Segmentation This is the official code for the arxiv. ZegFormer is the first framework that decouple the zero-shot seman





Jian Ding
 108  Dec 30, 2022









Code for CPM-2 Pre-Train


 CPM-2 Pre-Train Pre-train CPM-2 此分支为110亿非 MoE 模型的预训练代码，MoE 模型的预训练代码请切换到 moe 分支 CPM-2技术报告请参考link。 0 模型下载 请在智源资源下载页面进行申请，文件介绍如下： 文件名 描述 参数大小 100000.tar 





Tsinghua AI
 136  Dec 28, 2022









Unofficial PyTorch implementation of MobileViT.


 MobileViT Overview This is a PyTorch implementation of MobileViT specified in "MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Tr





Chin-Hsuan Wu
 348  Dec 23, 2022









A simple API wrapper for Discord interactions.


 Your ultimate Discord interactions library for discord.py. About | Installation | Examples | Discord | PyPI About What is discord-py-interactions? dis





james
 641  Jan 03, 2023









Open-sourcing the Slates Dataset for recommender systems research


 FINN.no Recommender Systems Slate Dataset This repository accompany the paper "Dynamic Slate Recommendation with Gated Recurrent Units and Thompson Sa





FINN.no
 48  Nov 28, 2022









:fire: 2D and 3D Face alignment library build using pytorch 


 Face Recognition Detect facial landmarks from Python using the world's most accurate face alignment network, capable of detecting points in both 2D an





Adrian Bulat
 6k  Dec 31, 2022









A simple code to convert image format and channel as well as resizing and renaming multiple images.


 Rename-Resize-and-convert-multiple-images A simple code to convert image format and channel as well as resizing and renaming multiple images. This cod





Happy N. Monday
 3  Feb 15, 2022









YoloAll is a collection of yolo all versions. you you use YoloAll to test yolov3/yolov5/yolox/yolo_fastest


 官方讨论群 QQ群：552703875 微信群：15158106211(先加作者微信，再邀请入群) YoloAll项目简介 YoloAll是一个将当前主流Yolo版本集成到同一个UI界面下的推理预测工具。可以迅速切换不同的yolo版本，并且可以针对图片，视频，摄像头码流进行实时推理，可以很方便，直观





DL-Practise
 244  Jan 01, 2023









Codebase for the solution that won first place and was awarded the most human-like agent in the 2021 NeurIPS Competition MineRL BASALT Challenge.


 KAIROS MineRL BASALT Codebase for the solution that won first place and was awarded the most human-like agent in the 2021 NeurIPS Competition MineRL B





Vinicius G. Goecks
 37  Oct 30, 2022









Lightweight tool to perform MITM attack on local network


 ARPSpy - A lightweight tool to perform MITM attack Using many library to perform ARP Spoof and auto-sniffing HTTP packet containing credential. (Never





MinhItachi
 8  Aug 28, 2022









This code is an implementation for Singing TTS.


 MLP Singer This code is an implementation for Singing TTS. The algorithm is based on the following papers: Tae, J., Kim, H., & Lee, Y. (2021). MLP Sin





Heejo You
 22  Dec 23, 2022









Official PyTorch implementation of our AAAI22 paper: TransMEF: A Transformer-Based Multi-Exposure Image Fusion Framework via Self-Supervised Multi-Task Learning. Code will be available soon.


 Official-PyTorch-Implementation-of-TransMEF Official PyTorch implementation of our AAAI22 paper: TransMEF: A Transformer-Based Multi-Exposure Image Fu






 117  Dec 27, 2022









Hooks for VCOCO


 Verbs in COCO (V-COCO) Dataset This repository hosts the Verbs in COCO (V-COCO) dataset and associated code to evaluate models for the Visual Semantic





Saurabh Gupta
 131  Nov 24, 2022









Code for “ACE-HGNN: Adaptive Curvature ExplorationHyperbolic Graph Neural Network”


 ACE-HGNN: Adaptive Curvature Exploration Hyperbolic Graph Neural Network This repository is the implementation of ACE-HGNN in PyTorch. Environment pyt






 9  Nov 28, 2022









Official implementation of DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations in TensorFlow 2


 DreamerPro Official implementation of DreamerPro: Reconstruction-Free Model-Based Reinforcement Learning with Prototypical Representations in TensorFl






 22  Nov 01, 2022









Code to reproduce the experiments in the paper "Transformer Based Multi-Source Domain Adaptation" (EMNLP 2020)


 Transformer Based Multi-Source Domain Adaptation Dustin Wright and Isabelle Augenstein To appear in EMNLP 2020. Read the preprint: https://arxiv.org/a





CopeNLU
 36  Dec 05, 2022