Improvement/234 create main evaluation script#243
Conversation
…tps://github.com/iai-group/UserSimCRS into improvement/233-create-classes-for-metrics
…tps://github.com/iai-group/UserSimCRS into improvement/233-create-classes-for-metrics
…mprovement/233-create-classes-for-metrics
…://github.com/iai-group/UserSimCRS into improvement/234-create-main-evaluation-script
…github.com/iai-group/UserSimCRS into improvement/234-create-main-evaluation-script
| annotate_dialogues(dialogues, user_nlu, agent_nlu) | ||
|
|
||
|
|
||
| def get_summary_by_agent( |
There was a problem hiding this comment.
We can’t skip calling it when there is only one agent because we won’t get a summary for it
NoB0
left a comment
There was a problem hiding this comment.
Some points need to be clarified
NoB0
left a comment
There was a problem hiding this comment.
For now LGTM (please check comments on confuse key access). Although, a follow-up should be created to investigate if build_metric_registry and evaluate_metric can be optimized with regards to maintenance when new metrics are supported.
| ) | ||
|
|
||
| if config["annotate_dialogues"].get(): | ||
| if not config["user_nlu"].get(None): |
There was a problem hiding this comment.
The get in confuse behaves bit differently than the get for a dictionary, i.e., it expects a template. I would suggest to look at the documentation, but I think that simply checking if the key is in the configuration and is not empty should work.
| raise ValueError( | ||
| "`user_nlu` is required when `annotate_dialogues` is True." | ||
| ) | ||
| if not config["agent_nlu"].get(None): |
No description provided.