Google Professional Machine Learning Engineer Question 214
Single answerGoogle Cloud PlatformYou are building a predictive maintenance model for a manufacturing company to detect equipment failures. The dataset includes time-series sensor data from multiple machines, and you need to choose an appropriate ML framework and model architecture. The requirements are: (1) Handle sequential data effectively, (2) Scale training across multiple GPUs on Google Cloud, and (3) Provide TensorFlow Extended (TFX) integration for production deployment. Which framework and model architecture should you choose?
- A
PyTorch with a Convolutional Neural Network (CNN)
- B
TensorFlow with a Recurrent Neural Network (RNN)
- C
Scikit-learn with a Random Forest
- D
TensorFlow with a Transformer-based architecture
Show answer and explanation
Correct answer: D
Explanation
The Transformer-based architecture is ideal for handling sequential data like time-series, as it overcomes the limitations of RNNs (e.g., vanishing gradients and sequential processing). TensorFlow provides robust multi-GPU training capabilities on Google Cloud and integrates with TensorFlow Extended (TFX) for production, making it the best choice for this use case.
- A. Incorrect.
PyTorch is a flexible framework, but it lacks built-in support for TensorFlow Extended (TFX), which is necessary for production deployment in this scenario. Additionally, CNNs are not ideal for sequential data like time-series.
- B. Incorrect.
TensorFlow supports TFX and scales well on Google Cloud, but RNNs are less efficient and harder to scale compared to modern architectures like Transformers.
- C. Incorrect.
Scikit-learn is not suitable for large-scale, GPU-accelerated training or handling sequential data effectively. It lacks the necessary scalability and integration with TFX.
- D. Correct.
TensorFlow with a Transformer-based architecture meets all the requirements: it handles sequential data effectively, supports distributed training on GPUs with TensorFlow, and integrates seamlessly with TFX for production deployment.