First NLP pipeline - ML Experiment: Train & Evaluate

Experiment - First NLP pipeline

Problem:Build a simple NLP pipeline to classify movie reviews as positive or negative.

Current Metrics:Training accuracy: 95%, Validation accuracy: 70%

Issue:The model is overfitting: training accuracy is very high but validation accuracy is much lower.

Your Task

Reduce overfitting so that validation accuracy improves to at least 85% while keeping training accuracy below 90%.

You can only modify the model architecture and training parameters.

Do not change the dataset or preprocessing steps.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

NLP

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, LSTM, Dense, Dropout
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.datasets import imdb

# Load data
max_features = 10000
max_len = 200
(X_train, y_train), (X_test, y_test) = imdb.load_data(num_words=max_features)

# Pad sequences
X_train = pad_sequences(X_train, maxlen=max_len)
X_test = pad_sequences(X_test, maxlen=max_len)

# Build model with dropout and smaller LSTM
model = Sequential([
    Embedding(max_features, 64, input_length=max_len),
    LSTM(32, return_sequences=False),
    Dropout(0.5),
    Dense(1, activation='sigmoid')
])

model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
              loss='binary_crossentropy',
              metrics=['accuracy'])

# Early stopping callback
early_stop = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)

# Train model
history = model.fit(X_train, y_train, epochs=20, batch_size=64, validation_split=0.2, callbacks=[early_stop])

# Evaluate on test data
loss, accuracy = model.evaluate(X_test, y_test)

print(f'Test accuracy: {accuracy * 100:.2f}%')

Added a Dropout layer with rate 0.5 after the LSTM layer to reduce overfitting.

Reduced LSTM units from 64 to 32 to lower model complexity.

Added EarlyStopping callback to stop training when validation loss stops improving.

Set learning rate to 0.001 for stable training.

Results Interpretation

Before: Training accuracy 95%, Validation accuracy 70% (overfitting)

After: Training accuracy 88%, Validation accuracy 86%, Test accuracy 85% (better generalization)

Adding dropout, reducing model size, and using early stopping help reduce overfitting and improve validation accuracy.

Bonus Experiment

Try using a 1D convolutional layer instead of LSTM for text classification and compare results.

💡 Hint

Replace the LSTM layer with Conv1D and MaxPooling1D layers, then train and evaluate.

Practice

1. What is the main purpose of an NLP pipeline in machine learning?

easy

A. To translate text into different languages automatically

B. To store large amounts of text data

C. To process text step-by-step for making predictions

D. To create images from text

Solution

1. Step 1: Understand the role of an NLP pipeline

   An NLP pipeline breaks down text processing into steps like cleaning, vectorizing, and modeling.

2. Step 2: Identify the goal of these steps

   The goal is to prepare text data so a model can make predictions, such as classifying or understanding text.

3. Final Answer:

   To process text step-by-step for making predictions -> Option C

4. Quick Check:

   NLP pipeline = step-by-step text processing for predictions [OK]

Hint: Remember: pipeline means step-by-step processing [OK]

Common Mistakes:

• Thinking pipeline stores data only
• Confusing pipeline with translation tools
• Assuming pipeline creates images

2. Which of the following is the correct way to import a text vectorizer from scikit-learn for an NLP pipeline?

easy

A. import CountVectorizer from sklearn.text

B. from sklearn.feature_extraction.text import CountVectorizer

C. from sklearn.vectorizer import TextCount

D. import text_vectorizer from sklearn.feature

Solution

1. Step 1: Recall the correct module for text vectorizers

   Scikit-learn provides CountVectorizer in the feature_extraction.text module.

2. Step 2: Check the import syntax

   The correct syntax is: from sklearn.feature_extraction.text import CountVectorizer.

3. Final Answer:

   from sklearn.feature_extraction.text import CountVectorizer -> Option B

4. Quick Check:

   Correct import = from sklearn.feature_extraction.text import CountVectorizer [OK]

Hint: Remember: CountVectorizer is in feature_extraction.text [OK]

Common Mistakes:

• Using wrong module names
• Incorrect import syntax
• Confusing class names

3. Given the following code snippet, what will be the output of print(X.toarray())?

from sklearn.feature_extraction.text import CountVectorizer
texts = ['cat and dog', 'dog and mouse']
vectorizer = CountVectorizer()
X = vectorizer.fit_transform(texts)
print(X.toarray())

medium

A. [[1 1 1 0]
[1 0 1 1]]

B. [[1 0 1 1]
[1 1 0 1]]

C. [[1 1 0 1]
[1 0 1 1]]

D. [[0 1 1 1]
[1 1 1 0]]

Solution

1. Step 1: Identify the vocabulary from the texts

   The texts are 'cat and dog' and 'dog and mouse'. The unique words are: 'and', 'cat', 'dog', 'mouse'. CountVectorizer sorts them alphabetically: ['and', 'cat', 'dog', 'mouse'].

2. Step 2: Map each text to counts of these words

   First text: 'cat and dog' -> counts: and=1, cat=1, dog=1, mouse=0 -> [1 1 1 0]. Second text: 'dog and mouse' -> counts: and=1, cat=0, dog=1, mouse=1 -> [1 0 1 1].

3. Final Answer:

   [[1 1 1 0] [1 0 1 1]] -> Option A

4. Quick Check:

   Vocabulary order and counts match [[1 1 1 0] [1 0 1 1]] [OK]

Hint: Remember: CountVectorizer sorts words alphabetically [OK]

Common Mistakes:

• Mixing word order in output
• Confusing counts of words
• Assuming different vocabulary order

4. You wrote this code but get an error: AttributeError: 'CountVectorizer' object has no attribute 'transform_text'. What is the likely fix?

from sklearn.feature_extraction.text import CountVectorizer
vectorizer = CountVectorizer()
vectorizer.transform_text(['hello world'])

medium

A. Replace transform_text with transform

B. Import CountVectorizer from a different module

C. Call fit before transform_text

D. Use fit_transform_text instead

Solution

1. Step 1: Identify the incorrect method name

   The error says 'CountVectorizer' has no method 'transform_text'. The correct method is 'transform'.

2. Step 2: Correct the method call

   Replace transform_text with transform to fix the error.

3. Final Answer:

   Replace transform_text with transform -> Option A

4. Quick Check:

   Correct method name is transform [OK]

Hint: Check method names carefully in docs [OK]

Common Mistakes:

• Using non-existent method names
• Not reading error messages
• Trying to call fit_transform_text which doesn't exist

5. You want to build a simple NLP pipeline that converts text to numbers and then trains a logistic regression model to classify text. Which sequence of steps is correct?

hard

A. Predict on new text -> Vectorize text -> Train logistic regression

B. Train logistic regression -> Vectorize text -> Predict on new text

C. Vectorize text -> Predict on new text -> Train logistic regression

D. Vectorize text -> Train logistic regression -> Predict on new text

Solution

1. Step 1: Understand the pipeline order

   First, text must be converted to numbers using vectorization before training a model.

2. Step 2: Follow logical flow

   After vectorizing, train the logistic regression model, then use it to predict on new vectorized text.

3. Final Answer:

   Vectorize text -> Train logistic regression -> Predict on new text -> Option D

4. Quick Check:

   Correct pipeline order = Vectorize text -> Train logistic regression -> Predict on new text [OK]

Hint: Always vectorize before training or predicting [OK]

Common Mistakes:

• Trying to train before vectorizing
• Predicting before training
• Skipping vectorization step