You need predictions on millions of records overnight. Which inference fits?
Why A is the answer
Batch
Batch inference suits large volumes where nobody is waiting.
In detail
- Large volumes that can wait suit batch inference.
- Choose real-time inference when a person or system is waiting.
Use case
An illustrative scenario, not a real incident.
A bank scores all its loans overnight. Nobody waits for a single answer, so batch inference fits.
See it in different ways
flowchart TD
A{"When is the result needed?"} -->|"Immediately"| B["Real-time"]
A -->|"Soon, for large payloads"| C["Asynchronous"]
A -->|"Later, for many records"| D["Batch"]stateDiagram-v2 [*] --> Data Data --> Train Train --> Evaluate Evaluate --> Deploy Deploy --> Monitor Monitor --> Data: drift
flowchart TD
Q{"When is the result needed?"}
Q -->|"Right now"| A0["Real-time inference"]
Q -->|"Later"| A1["Batch inference"]Key point: Batch suits large volumes with no one waiting.
Why the other options are wrong
- B. Real-time
- Real-time is for immediate answers.
- C. Serverless, for each single user request
- That is built for single requests.
- D. A human review of each record
- People cannot review millions of records.