OpenMined/pygrid-admin

Run worker

dachirui opened this issue · 6 comments

Question

How can I run/connect a worker with the domain and see the worker on http://localhost:3000/infrastructure (worker tab)?

Hi @dachirui ! Thanks for your question. I suggest you take a look at the latest Jupiter notebooks on how to run computations on remote workers, and launch them on a cloud provider here. Regarding the infrastructure tab, I believe this is broken rn and we're working to get it fixed asap.

Thank you very much! It works until I try to create a dataset on the owner side provided here
When I run dataset = domain.datasets.create('diabetes-dataset.tar.gz') it throws
`---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
in
----> 1 dataset = domain.datasets.create('diabetes-dataset.tar.gz')
2 dataset.pandas

AttributeError: 'GridClient' object has no attribute 'datasets'`

Is there any other attribute to create a dataset?

Thanks

@dachirui The code that implements those methods only lives inside the pygrid_demo branch. You will have to reinstall PySyft from that branch pip3 install git+https://github.com/OpenMined/PySyft.git@pygrid_demo so those methods are accessible.

@victorperezc thank you very much, was ok until dataset creation where same line throws JSONDecodeError: Expecting value: line 1 column 1 (char 0)
I am looking for it but it could be a problem related with Python version. I am using 3.8.

Yeah this can happen because of a variety of things. The endpoint that accepts datasets is very sensible and only accepts specific formats. For instance, try to make sure that your dataset follows this

  • It's a compressed directory with extension .gz.zip
  • The compressed directory contains .csv files and a manifest, tags and description files.
  • Make sure the csv files are comma separated

Since this matter is out of the scope of this repository I suggest you either open an Issue in PyGrid or send me a message over Slack. I will be happy to help.

Closing the Issue now ~