Rasa: Understanding the confidence level

[oximer]

I'm executing the /data/example/rasa example as described on tutorial.
I'm trying to understand why I'm not getting 100% of confidence when I send the same text that I used on during the training.

curl -XPOST localhost:5000/parse -d '{"q":"I am looking for mexican indian fusion"}' | python -mjson.tool
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   364  100   318  100    46   4694    679 --:--:-- --:--:-- --:--:--  4676
{
    "entities": [
        {
            "end": 38,
            "entity": "cuisine",
            "extractor": "ner_mitie",
            "start": 17,
            "value": "mexican indian fusion"
        }
    ],
    "intent": {
        "confidence": 0.9046218833485684,
        "name": "restaurant_search"
    },
    "text": "I am looking for mexican indian fusion"
}

The confidence of it should be 1, right? Why it's 0,9?
I have trained it which the exact same words....

When using Wit or API, I get 1 for those scenario.
What I'm doing wrong?

Thanks

[PHLF]

To quote @wrathagom from GItter:

It's not just a straight lookup, so no the confidence shouldn't be 1.

Don't try to compare too much RASA with others services from a technical point of view: if they follow similar core ideas, they're not backed by the same technologies so you can only expect them to perform differently on the same dataset.

I really advise you to read @amn41 blog post : it will give you better insights about how everything works.

[cuddlex]

I have a similar question about this. Is it possible to have a balance confidence? I mean if a entity appears in two different intents, the confidence will be 50% per each intent? In my training data, in each intent there is the same entity example number but the model always gives a high confidence for one intent.

[tmbo]

The intent classification doesn't rely on the entities present - they are two separate things.

[wrathagom]

@oximer I'm closing because I think you're question has been answered, if not let us know.