Passing variables from one experiment to the next

[AlvaroAguilera]

Hello,

is it possible to pass some intermediate results / variable from one test to the next one when working headless?
How about stopping the test chain / marking the token as finished according to some criteria before the normal end?

Thanks

[vsoch]

Are you looking for the experiment factor containerized version? This is for expfactory.org, which doesn't have the concept of headless.

[AlvaroAguilera]

Yes. I wasn't aware of the difference, sorry. Where is the best place to ask this question?

[vsoch]

I can transfer for you over to the expfactory repo!

[vsoch]

okay all set!

is it possible to pass some intermediate results / variable from one test to the next one when working headless?

So to answer your question, the expfactory here does support experiment variables that you could set up for an experiment beforehand, but we've never had moving them between experiments, which (imho) goes against the modular philosophy of expfactory, that each is a moveable experiment, because it would not make them different experiments. Have you thought about putting the experiments together that have some dependency on one another?

How about stopping the test chain / marking the token as finished according to some criteria before the normal end?

This would also be done via some logic in your javascript to end the experiment early and then progress to the end. Is the "between experiment" variable you have for this use case?

[AlvaroAguilera]

For our studies, we are using some very long tests (2 hours+) that are error prone and difficult to debug. That's why we want to divide these tests into smaller steps but for that we need a way to pass the intermediate results to keep the logic working.

Jumping to the finish would be an option, while not the ideal one. Is the token then marked as finished, when we skip some of the tests?

[vsoch]

For our studies, we are using some very long tests (2 hours+) that are error prone and difficult to debug. That's why we want to divide these tests into smaller steps but for that we need a way to pass the intermediate results to keep the logic working.

Wow that's long - and to break into smaller experiments you need the data from the previous (so is it worth it if the data is the same large siez)? For saving data, you can continue to post to /save to do a save for the current state. And at any point you can hit /finish to finish an experiment, e.g., see https://expfactory.github.io/contribute#experiment-pre-reqs. My first inclination is that the experiment is too long and should be broken up, and try to remove the dependency between the experiments. But it does sound like this is an interesting use case to pass data between experiments, or just send a message to the server to "finish the whole thing." If you want to spec out a design for that, and either work on it and PR or wait for me to have a weekend to do so, I could help. This isn't in any way related to my work so I couldn't jump on it immediately.

[vsoch]

E.g., we could make another endpoint for your experiment to ping to pass some set of variables, and they would be added to the experiment variables that we currently have, and then passed to the directed experiment(s) they are intended for, and that experiment can choose to use them (or ignore). If a variable is repeated from the original set it would be re-written. Then what you could do is split your long experiment into different phases and pass forward whatever you need.

[vsoch]

Is this still an open question / issue?