I think the latter. the specific use-case I'm talking about is running experiments on one machine, and using a local server on another machine to read the "logs" \ artifacts
Hi RipeGoose2 ,
Are you talking about the experiment output (as in stdout
) or specifically the artifacts added using Task.upload_artifact()
?
The artifacts are reported and uploaded to the server when the experiment is running, and you can view them there when the experiment is running and also after the experiment has completed.
What is your specific scenario? Is there any specific reason you don't want the artifacts to be reported/uploaded during the experiment's run?
But that machine doesn't have to run the server - the server can run on a different machine (even your own machine). Moreover, the trains package you import in your experiment code requires access to the Trains Server, and does not support running with no server connection.
Just to be clear, when you say "artifacts", do you refer also to experiment metrics (scalars etc.), or just to artifact "objects" (like files, numpy arrays, pandas dataframes etc. that are reported using Task.upload_artifact()
)?
yes, I have limited access to the machine that is running the experiment. I can't setup a server there. but I want to collect the results and view them later
We are thinking about an offline mode in which all the experiment results are stored locally and then reported back to the server when a connection is available, however this is not yet available in Trains...
If that's of interest to you, I'll appreciate if you open an issue with a feature request in the trains GitHub page 🙂
I refer to all the info that accessible through the webApp
"does not support running with no server connection." this is what I was afraid of..I'll need to figure out if I can use trains at all 😞
If you just need to access the artifacts uploaded by a remotely-executed experiment, you can just go to the experiment's page in the WebApp and look in the Artifacts tab - each artifact there has both a preview, and a download link to the artifact file uploaded to the trains files server.
by communication that the artifacts are streamed from the machine running the experiments to the local server?
can it be done "offline" after the experiments run view them in my local server?
if I don't have internet connection on the other machine, can I just copy the artifacts and transfer them to my local machine?
by WebApp you mean the public online one? I might be confusing stuff
and I will also be happy to see if I can contribute maybe to this specific feature or maybe others
Hi RipeGoose2 all PR's are welcome, feel free to submit :)
yes I will be happy to, its gonna be my first time
RipeGoose2 , I assume you're running your own local Trains Server, and that the machine running the remote experiment (using Trains Agent) is communicating with your local server. By WebApp I mean the web interface served by your local server (http://<server-ip>:8080)