Self-contained python library

[ericjeangirard]

Hi!
The current Grobid Python package (https://github.com/kermitt2/grobid_client_python) is a wrapper for the actual Grobid API, which must be deployed somewhere.
In come context, having a full self-contained python library could be pretty handy. A simple pip install pygrobid without having to bother with docker or java would be great.

[lfoppiano]

Thanks @ericjeangirard. I moved the issue under this repository so that we can plan the changes:

1. should remove the config.json file which is mostly useless
2. defaulting on one or a list of grobid services deployed on huggingface
3. add a big warning saying that for serious processing it's better to set up a larger instance

The free huggingface services can be used with the CRF images processing also a larger batch (100-1000 documents) without problems. However, the full image will need to be limited to one article at the time, unless we want to have something more scalable, but then it should be a paid service.

Do you think that asking for an API_KEY registered to a credit card would be accepted by people?

[ericjeangirard]

Thanks @lfoppiano !
For the paying API_KEY, I have no clue to be honest.
Also, in some environments, once the requested packages have been installed, internet access is not allowed. It's typically the case in some industry (esp. military), but also in the context of some super-computer infrastructure. Not to mention some Kaggle competions!

[lfoppiano]

We should also remove the need of the config file, which can be the default

[lfoppiano]

@ericjeangirard I've partially implemented the change. By default the client will hit localhost, and we added a --server parameter, together with the demo services, should make the process much easier.