Giovanni Campagna

This issue was transferred from almond-dialog-agent. It should be implemented in terms of templates: we change the templates such that filters and imprecise answers can use numbers where measurements are...

Underlying issue: https://github.com/mapbox/node-sqlite3/issues/1381

We should use [howl](https://github.com/castorini/howl) to train the wakeword and [honkling](https://github.com/castorini/honkling) (embedded howl with Tensorflow.js + WASM) at inference time.

We don't have yet a template or dialogue act to do "more results".

The problem is that once you merge an API, you have to support it forever (until the next major API break). You have to compare the small amount of work...

Also "list my devices"

This is literally a bug in the tokenizer, right? We can fix it without significant changes.

Are you going to fix these any time soon?

Other log: ``` ==== # main/548e61fd-88ad-444a-a3a3-5c7160a80ac9 #! timestamp: 2021-07-23T20:51:51.596Z U: tell me a joke UT: $dialogue @org.thingpedia.dialogue.transaction.execute; UT: @com.icanhazdadjoke.get(); C: $dialogue @org.thingpedia.dialogue.transaction.execute; C: @com.icanhazdadjoke.get() C: #[results=[ C: { id="4wciyk3EBAd"^^com.icanhazdadjoke:id, text="I...

This seems to work with the latest model, given the fix we did to future dates. Closing.