Skip to content
This repository was archived by the owner on May 1, 2025. It is now read-only.
This repository was archived by the owner on May 1, 2025. It is now read-only.

Training on custom dataset and answer is always "o" #49

@ashleyyy94

Description

@ashleyyy94

I am trying to train on a custom WOZ dataset. I have 2 jsonl files, train.jsonl and val.jsonl as suggested.

A sample line from my dataset is {"context": "no, i just need to make sure it's cheap. oh, and i need parking", "question": "What is the change in state?", "answer": "name: not mentioned, area: not mentioned, parking: yes, pricerange: cheap, stars: not mentioned, internet: not mentioned, type: hotel;"}

This should be the correct format for the dataset training tuple. The problem is when I run the training process, when it tries to run validation after 1000 iterations (--val_every 1000), the output for the answer is always "o". As shown below:

greedy: 'food: chinese, pricerange: expensive, name: not mentioned, area: centre;'

answer: 'o'

context: 'how about the centre?'

question: 'what is the change in state?'

May I know what I'm doing wrongly? The wrong answer "o" is causing all the metrics like joint goal em, nf1 and nem to be zero.

Thank you.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions