How to actually reproduce an experiement

[wheatdog]

Sorry, I am new to sacred. I am not sure whether this is the right place to ask.

I know we can use Observers to record the information of experiments. But I am wondering how to use those data to actually reproduce an experiment? Do we have to manually apply those configs and codes?

Thanks

[Qwlouse]

Yes, unfortunately, there is currently no automated process for reproducing experiments. You need to get the right source code (possibly from the database), and the config from the database and then run your experiment with ex.run(config_updatas=config_from_db). I'm working on an automated way of doing that, but I'm rather busy, so it is going to take a while.

[Leinadj]

I have written a solution for myself, but I am not sure what a general solution would need to look like. The way I am doing it is to create a virtual environment directly from python, install all the required packages, load all the source code files from the database and then run the main source code file. The main problem in my opinion is the storage of the data (if its quite big for example), which should be stored a the very same place as it was when the experiment was run initially. Any thoughts on this topic? I would appreciate help, so that I can maybe try to generalize my solution and fit it into the sacred library.

[wheatdog]

@Leinadj Can you share your code somewhere? Although I am not using sacred now, I am still willing to see it!