Model Card for Model ID
Downloads
!pip install contractions
!pip install textsearch
!pip install tqdm
import nltk
nltk.download('punkt')
Fundamental classes
import tensorflow as tf
from tensorflow import keras
import pandas as pd
import numpy as np
Time
import time
import datetime
Preprocessing
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing import sequence
from sklearn.preprocessing import LabelEncoder
import contractions
from bs4 import BeautifulSoup
import re
import tqdm
import unicodedata
seed = 3541
np.random.seed(seed)
Define a dummy loss to bypass the error during model loading
def dummy_loss(y_true, y_pred):
return tf.reduce_mean(y_pred - y_true)
Loading the model Trained on Yelp reviews
modelYelp = keras.models.load_model(
'/kaggle/input/pre-trained-model-binary-cnn-nlp-yelpreviews/tensorflow1/pre-trained-model-binary-cnn-nlp-yelp-reviews/1/Binary_Classification_90_Yelp_Reviews_CNN.h5',
compile=False
)
Compile the model with the correct loss function and reduction
modelYelp.compile(
optimizer='adam',
loss=keras.losses.BinaryCrossentropy(reduction=tf.keras.losses.Reduction.SUM_OVER_BATCH_SIZE),
metrics=['accuracy']
)
Loading Yelp test data
dataset_test_Yelp = pd.read_csv('/kaggle/input/yelp-reviews-for-sentianalysis-binary-np-csv/yelp_review_sa_binary_csv/test.csv')
Loading Yelp train data (to be used on the label encoder)
dataset_train_Yelp = pd.read_csv('/kaggle/input/yelp-reviews-for-sentianalysis-binary-np-csv/yelp_review_sa_binary_csv/train.csv')
Shuffling the Test Data
test_Yelp = dataset_test_Yelp.sample(frac=1)
train_Yelp = dataset_train_Yelp.sample(frac=1)
Taking a tiny portion of the database (because it will only be used on the label encoder)
train_Yelp = dataset_train_Yelp.iloc[:100, :]
Taking only necessary columns
y_test_Yelp = test_Yelp['class_index'].values
X_train_Yelp = train_Yelp['review_text'].values
y_train_Yelp = train_Yelp['class_index'].values
Preprocess corpus function
def pre_process_corpus(corpus):
processed_corpus = []
for doc in tqdm.tqdm(corpus):
doc = contractions.fix(doc)
doc = BeautifulSoup(doc, "html.parser").get_text()
doc = unicodedata.normalize('NFKD', doc).encode('ascii', 'ignore').decode('utf-8', 'ignore')
doc = re.sub(r'[^a-zA-Z\s]', '', doc, re.I|re.A)
doc = doc.lower()
doc = doc.strip()
processed_corpus.append(doc)
return processed_corpus
Preprocessing the Data
X_test_Yelp = pre_process_corpus(test_Yelp['review_text'].values)
X_train_Yelp = pre_process_corpus(X_train_Yelp)
Creating and Fitting the Tokenizer
etc ...
More info on the Model page on Kaggle :
