search

bluesky / ophyd

hardware abstraction in Python with an emphasis on EPICS
https://blueskyproject.io/ophyd
BSD 3-Clause "New" or "Revised" License
49 stars 78 forks source link

EpicsSignal based on multiple signals #486

Closed jrmlhermitte closed 4 years ago

jrmlhermitte commented 6 years ago

Say you have a signal made of multiple signals:

sig1 = EpicsSignalRO(pvname1, name=name1)
sig2 = EpicsSignalRO(pvname1, name=name1)

how to create a new signal such that the result is sig1.value+sig2.value - bg where bg is some precomputed parameter?

cowanml commented 6 years ago

On Fri, 17 Nov 2017, Julien Lhermitte wrote:

Say you have a signal made of multiple signals:

sig1 = EpicsSignalRO(pvname1, name=name1) sig2 = EpicsSignalRO(pvname1, name=name1)

how to create a new signal such that the result is sig1.value+sig2.value - bg where bg is some precomputed parameter?

If that new signal isn't a one-off one time use thing, you're best off pushing stuff like that down into epics so it's available like a regular pv to any pv client.

-matt

danielballan commented 6 years ago

There are a couple considerations here:

  1. Is the computation established and stable, or is it part of an experimental workflow that still requires some improvisation? Doing this in the Python layer is much better for interactive exploration.
  2. Is this something that is straightforward to do in C, or would numpy/scipy/scikit-image/scikit-learn help?
  3. Is this something that other users at the facility would want to check or monitor? Consider EPICS!

To your question, ophyd provides DerivedSignal and AttributeSignal. I can never remember which is which! Better documentation is coming. For now, grep the test suite.

cowanml commented 6 years ago

On Fri, 17 Nov 2017, Dan Allan wrote: ...

  1. Is the computation established and stable, or is it part of an experimental workflow that still requires some improvisation? Doing this in the Python layer is much better interactive exploration.

sure, although I worry about temporary solutions becoming standards. You can do similar stuff in css and I tell people the same thing.

  1. Is this something that is straightforward to do in C, or would numpy/scipy/scikit-image/scikit-learn help?

If it's the simple example mentioned, pv1+pv2-pv3, no c required, simple epics calc record.

  1. Is this something that other users at the facility would want to check or monitor? Consider EPICS!

Or that you might want in another process or css or whatev, doesn't have to be for other people to bother with epics.

I'm not trying to preach epics, just reminding it's there, has some nifty capabilities/benefits, and isn't always painful.

-matt

danielballan commented 6 years ago

Thanks for that useful perspective, @cowanml.

jrmlhermitte commented 6 years ago

yes thanks @cowanml this reminder is appreciated, please do keep giving feedback! :-)

jrmlhermitte commented 6 years ago

PS: I will contact the controls group on such an instance and close this issue once resolved. In the meantime, a pseudo-example (with a big disclaimer to contact the controls group) could be something to PR, I'll work on that later.

tacaswell commented 4 years ago

Closing as stale.

If we really need this at the python level this is a pseudo-positioner, but it should be done in the IOC when ever possible.