Google Professional Machine Learning Engineer Question 515
Single answerGoogle Cloud PlatformYou are designing a machine learning pipeline on Google Cloud to classify customer reviews into positive and negative sentiments. The dataset contains text data with thousands of reviews. After preprocessing, you decide to use embeddings to represent the text data. However, the embedding vectors have a high dimension, which could cause computational inefficiencies. What is the best approach to reduce the dimension of the embeddings while ensuring minimal loss of information?
- A
Apply Principal Component Analysis (PCA) to reduce the dimensionality of the embeddings.
- B
Use feature hashing to randomly project the embeddings into a lower-dimensional space.
- C
Increase the number of training samples to improve the quality of the embeddings.
- D
Use t-SNE (t-Distributed Stochastic Neighbor Embedding) to reduce the dimensionality of the embeddings.
Show answer and explanation
Correct answer: A
Explanation
Principal Component Analysis (PCA) is the best option for reducing the dimensionality of embeddings in this scenario. It effectively reduces the number of dimensions while retaining the most important features of the data, making it suitable for machine learning tasks. Other options either do not address the problem of dimensionality directly or are not practical for this use case.
- A. Correct.
Principal Component Analysis (PCA) is a widely-used technique for dimensionality reduction that projects data into a lower-dimensional space while retaining as much variance as possible. It is computationally efficient for high-dimensional data like embeddings.
- B. Incorrect.
Feature hashing reduces dimensionality by randomly projecting data, but it may lead to information loss due to collisions, and it is not ideal for embeddings where preserving relationships between dimensions is critical.
- C. Incorrect.
Increasing the number of training samples does not directly address the issue of high dimensionality. It may improve the quality of the embeddings but will not reduce their dimensionality.
- D. Incorrect.
t-SNE is primarily used for visualization of high-dimensional data in 2D or 3D. It is computationally expensive and not suitable for preprocessing embeddings for machine learning pipelines.