LDA with Gensim in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of LDA Topic Distribution

Given the following code snippet using Gensim's LDA model, what is the output of the print statement?

NLP

import gensim
from gensim import corpora

texts = [['apple', 'banana', 'apple', 'fruit'], ['banana', 'fruit', 'fruit', 'apple'], ['car', 'engine', 'wheel'], ['engine', 'car', 'wheel', 'car']]
dictionary = corpora.Dictionary(texts)
corpus = [dictionary.doc2bow(text) for text in texts]

lda = gensim.models.LdaModel(corpus, num_topics=2, id2word=dictionary, random_state=42, passes=10)

print(lda[corpus[0]])

A [(0, 0.05), (1, 0.95)]

B [(0, 0.5), (1, 0.5)]

C [(0, 0.95), (1, 0.05)]

D [(0, 1.0), (1, 0.0)]

💡 Hint

Think about how LDA assigns topic probabilities to documents based on word distributions.

The first document mostly contains words related to the first topic, so the model assigns a high probability to topic 0 and a low probability to topic 1.

❓ Model Choice

intermediate

Choosing Number of Topics in LDA

You want to model topics in a collection of news articles using Gensim's LDA. Which approach best helps decide the number of topics?

A Use domain knowledge to guess a reasonable number and validate with coherence scores.

B Set num_topics to a very high number and pick the one with the highest coherence score.

C Always set num_topics to 2 for simplicity.

D Use the number of documents as num_topics.

💡 Hint

Think about balancing model complexity and interpretability.

Choosing num_topics based on domain knowledge and validating with coherence scores helps find a meaningful and interpretable number of topics.

❓ Hyperparameter

advanced

Effect of passes Parameter in Gensim LDA

What is the effect of increasing the passes parameter when training an LDA model with Gensim?

A It increases the number of times the model iterates over the entire corpus, potentially improving convergence.

B It changes the number of topics the model will find.

C It controls the number of words in each topic.

D It sets the random seed for reproducibility.

💡 Hint

Think about how many times the model sees the data during training.

The passes parameter controls how many times the model goes through the whole corpus during training, which can help the model converge better.

❓ Metrics

advanced

Interpreting Coherence Score in LDA

You trained two LDA models with different numbers of topics. Model A has a coherence score of 0.42, and Model B has 0.58. What does this imply?

A Coherence scores do not relate to topic quality.

B Model A has better topic diversity than Model B.

C Model A is overfitting the data compared to Model B.

D Model B's topics are more semantically coherent and likely more interpretable.

💡 Hint

Higher coherence scores usually mean better topic quality.

Coherence measures how semantically related the top words in topics are. Higher scores indicate more meaningful topics.

🔧 Debug

expert

Identifying Error in LDA Corpus Preparation

What error will this code raise when running the LDA model training?

NLP

import gensim
from gensim import corpora

texts = [['dog', 'cat'], ['cat', 'mouse'], ['dog', 'mouse']]
dictionary = corpora.Dictionary(texts)
corpus = [dictionary.doc2bow(text) for text in texts]

# Incorrect corpus: passing list of words instead of bow vectors
lda = gensim.models.LdaModel(texts, num_topics=2, id2word=dictionary, passes=5)

A TypeError: unhashable type: 'list'

B AttributeError: 'list' object has no attribute 'get'

C No error, model trains successfully

D ValueError: corpus must be a list of bag-of-words vectors

💡 Hint

Check what type the LDA model expects as corpus input.

The LDA model expects a corpus of bag-of-words vectors, but the code passes raw token lists, causing an AttributeError when the model tries to access methods on the corpus elements.