Skip to main content

Classification Patterns

Classification is one of the most common LLM use cases. BAML makes it reliable with enum types.

Basic Sentiment Classification​

enum Sentiment {
  positive @description("Happy, satisfied, enthusiastic")
  negative @description("Unhappy, frustrated, disappointed")
  neutral @description("Neither positive nor negative")
}

function ClassifySentiment(text: string) -> Sentiment {
  client GPT4
  prompt #"
    Analyze the sentiment of this text:
    {{ text }}

    Classify as positive, negative, or neutral.
  "#
}

Multi-Label Classification​

When content can have multiple labels:

enum Topic {
  technology
  business
  politics
  entertainment
  sports
}

function ClassifyTopics(article: string) -> Topic[] {
  client GPT4
  prompt #"
    What topics does this article cover?
    {{ article }}

    List all applicable topics.
  "#
}

Classification with Confidence​

class ClassificationResult {
  label Sentiment
  confidence float @description("0.0 to 1.0")
  reasoning string
}

function ClassifyWithConfidence(text: string) -> ClassificationResult {
  client GPT4
  prompt #"
    Classify this text and explain your reasoning:
    {{ text }}
  "#
}

Tips​

  1. Use @description to guide the model on edge cases
  2. Be specific in enum values - avoid ambiguous labels
  3. Add reasoning when you need to debug classifications

Concepts​