A flood desk brags that an adapted model has lower perplexity on river chatter, then ships it as a flood / no-flood classifier without scoring those labels. What did they mix up?
Select an answer to reveal the explanation.
Short Explanation
Lower perplexity on river chatter is a better language-model fit, not a flood / no-flood trophy. Ship the labels only after a classification score. The two numbers do not always move together, and a summary overlap on the chatter is equally off-task.
Full Explanation
Perplexity measures how well a language model fits a stream of text; lower is a better fit, not a better classifier. Flood versus no-flood is a labeled task and needs its own score, such as accuracy or rare-label F1. A perplexity drop on river chatter can happen while the emergency labels stay wrong. Summary overlap on the chatter is equally off-task.