Search⌘ K
AI Features
Home
Courses
Building Advanced Deep Learning and NLP Projects

Assignment Solution

Explore building an RNN model for IMDB reviews sentiment analysis by increasing vocabulary size, tuning hyperparameters, and predicting sentiment for new sentences using word-to-integer mappings and embeddings.

We'll cover the following...

Problem one

You’ll need to create the model with more words. In the project, we only used 10,000 words, so increase this number.

Problem two

Try to change the hyperparameters and increase the accuracy of the model.

Below is the code snippet for the above two problem statements.

Python 3.5
# Import the required libraries
from tensorflow.keras.datasets import imdb
from tensorflow.python.keras.callbacks import ModelCheckpoint
from tensorflow.python.keras.callbacks import EarlyStopping
from tensorflow.python.keras.preprocessing import sequence
from tensorflow.python.keras.layers import Embedding,SimpleRNN,Dense
from tensorflow.python.keras.models import Sequential
# Load the Dataset
((XT,YT),(Xt,Yt)) = imdb.load_data(num_words=30000)
print("The length of the Training Dataset is ", len(XT))
print("The length of the Testing Dataset is ", len(Xt))
# Perform the padding
X_train = sequence.pad_sequences(XT,maxlen=500)
X_test = sequence.pad_sequences(Xt,maxlen=500)
# Create the Model Architecture
model = Sequential()
model.add(Embedding(30000,128))
model.add(SimpleRNN(64))
model.add(Dense(1,activation='sigmoid'))
# Compile the Model
model.compile(optimizer='rmsprop',loss='binary_crossentropy',metrics=['acc'])
# Create the Callbacks
checkpoint = ModelCheckpoint("best_model.h5", monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=False)
earlystop = EarlyStopping(monitor='val_acc',patience=1)
# Train the Model
hist = model.fit(X_train,YT,validation_split=0.2,epochs=10,batch_size=128,callbacks=[checkpoint,earlystop])
# Evaluate the Model on Test Dataset
model.evaluate(X_test,Yt)

Explanation:

  • We used a similar code which we discussed in the previous lessons.
  • On line 10, we used the top 30000 frequent
...