stanford-oval / genie-parser Goto Github PK

We should provide reference docker files for easy deployment.

Additionally, we could use docker ourselves to be able to run on unmodified training machines, using a container management system.

Hello,
Could you upload some example dataset? There should be only simple sentences or should have some specific format?
regards

While bootstrapping a new language, you have no model you can load. To get a model, you need a dataset, and to get a dataset you need to write a dataset.tt file in Thingpedia - at which point Thingpedia will call genie-parser to tokenize the sentences written into it and you'll be sad.

Depends on #5 to do something sensible if somebody throws an invalid language at us.

Loading the models from the configuration file does not work, if one wants to add & remove models dynamically from Thingpedia.

Right now, if the tokenizer fails with an error, we crash trying to read "tokens" from a response that won't have it.