Random Candidate

This is an agent which chooses a random label candidate, to be used for debugging or as an extremely simple baseline.

If label candidates are available for the task, it chooses one randomly as a response. It randomly sorts the candidates to return in the text_candidates field, used for calculating hits@k metrics. If label candidates are not available, it just repeats the label.

Basic Examples

Evaluate on bAbi by choosing a random candidate.

parlai eval_model -m random_candidate -t babi

RandomCandidateAgent Options

RandomCandidateAgent Arguments

Argument

Description

--label-candidates-file

File of candidate responses to choose from