PySyft
PySyft copied to clipboard
Federated Learning example with syft version 0.7
Hi,
I have been trying to find a federated learning tutorial with the latest syft version, but all documents are related to older versions, which includes a bunch of function that are not supported by version 0.7. I want to use PySyft for my research but no help so far.
Hi @bussfromspace - we're working very hard to bring FL back on the new foundation. It will be possible in 0.8 (close to dev complete, testing happening in march for a release at the end of the month) but a bit inconvenient. It's looking like 0.9 will have robust support for FL.
When using pysyft version 0.7, I faced the same problem. I was excited when I first encountered the concept of federated learning, but every time I tried to run the code, I became frustrated with the disconnect between the tutorials and the code, as well as the outdated code version.
UPDATE: We've made some tweaks to final scope and it looks likely there'll be a FL example in 0.8. Please give us a couple weeks to work out the kinks.
When using pysyft version 0.7, I faced the same problem. I was excited when I first encountered the concept of federated learning, but every time I tried to run the code, I became frustrated with the disconnect between the tutorials and the code, as well as the outdated code version.
So sorry for the frustration. We've had to take several steps backwards to ultimately find the right path forward. I hope you'll find the wait was worth it but I suppose we won't know that until it's released. Thank you for your patience and we'll increase our vigilence w.r.t deleting old documentation.
Hi,
Unfortunately when I code with syft version 0.7 I cannot use hook or torch hook, as I understood in the new version some ways of coding have been changed, but unfortunately I could not find any documentations helping me to understand how do I have to code. I just saw that I must use only VirtualMachine but further I encounter again some problems because of changes. Let me know please from where can I see how do I have to code in syft version 0.7
Thanks
Any update on this topic?
You can do federated learning by using the 00-04 tutorials here (https://github.com/OpenMined/PySyft/tree/dev/notebooks/api/0.8), but we haven't made an end-to-end example tutorial. We're currently interviewing candidates who would do so if/when they are hired.
Thank you for your patience in this area. For the last two years we've prioritized figuring out how to really nail access to data at one institution (in particular overcoming legal, security, privacy, ethics, PR, trust & safety and other issues which tend to block deployment of PETs). A solution to working with data at one institution does generalize to federated learning without too much fuss, but we haven't focused on that generalization in a while. However, I think we're getting to the point where access to data at one institution is getting easy enough that we could start thinking about FL. Join the slack (slack.openmined.org) to get a notification when we do (i.e. new tutorials).
Closing for now.