Production Machine Learning Systems

Question 1

What is static and dynamic prediction?

Answer 1

• Static prediction - caculate prediction in batches, store it to databse and take the results of a prediction from a database on request
• Dynamic prediction - calculate prediction on the fly

Question 2

What is Peakedness and Cardianality and how it is realated to choosing between static and dynamic model serving?

Answer 2

Peakedness - how much is data concentrated aroud the mean
Cardinality - number of values in the set

Question 3

What are the options to make your training faster?

Answer 3

1. Use a more powerful device
2. Optimize your input pipeline
3. Distributed training

Question 4

What is a distributed training?

Answer 4

Splitting your training workload across multiple worker nodes (potentially with multiple GPU each) connected over a network