Feature request: to_glue #153

keflavich · 2014-10-09T07:52:56Z

@ChrisBeaumont this is for you, maybe @astrofrog too since you made the successful one-off pvextractor widget.

Would it be possible to open a cube in Glue, or at least with the Glue quick-viewer, from within an existing session? I'm thinking of getting access to on-the-fly coordinate changes within glue, essentially, but starting from a regular old command line application.

astrofrog · 2014-10-08T14:13:01Z

I've been thinking that the standalone PV extractor I put together could in fact be a cube viewer (so also have a panel for the spectral extractor) and so to_glue could open this up?

keflavich · 2014-10-08T14:36:23Z

That's basically what I had in mind. I'll work on it if I finish real work today... =)

ChrisBeaumont · 2014-10-08T15:13:33Z

I'll work on it if I finish real work today... =)
ouch :)

There are several ways to start glue from the command line. The easiest is qglue http://www.glueviz.org/en/stable/glue_from_python.html#quickly-send-data-to-glue-with-qglue , though it doesn't let you pre-configure viewers. If you want to automatically add a cube view, you have to use something like this: http://www.glueviz.org/en/stable/data_viewer_options.html#programmatically-configuring-plots

keflavich · 2014-10-09T08:11:35Z

@astrofrog @ChrisBeaumont WIP. It works so far in terms of sending data to a Glue application, but it has lots of problems:

the spectral axis is not WCS, even though the spatial axes are
there are a lot of warnings and logging messages popping up
it's not obvious how to send multiple cubes to the same glue app in a nice way - would be best to retrieve the current glue session if possible (the global one as opposed to the SpectralCube._glue_app)
Need to auto-open the image viewer

keflavich · 2014-10-09T08:55:14Z

spectral_cube/spectral_cube.py

+        if data_collection is not None:
+            if name in data_collection:
+                name = name+"_"
+            data_collection[name] = result


This is wrong, but it was based off of http://www.glueviz.org/en/stable/glue_from_python.html#usage-examples. I guess catalog is not a DataCollection in that example but something else.

ChrisBeaumont · 2014-10-09T15:06:31Z

the spectral axis is not WCS, even though the spatial axes are

Can you elaborate on that -- is the problem that the cube is weird, or that Glue isn't displaying something properly, or...?

there are a lot of warnings and logging messages popping up

I'll take a look. Glue generates lots of logging messages anyways, so it might be benign, and we just need to send them someplace more quietly

it's not obvious how to send multiple cubes to the same glue app in a nice way - would be best to retrieve the current glue session if possible (the global one as opposed to the SpectralCube._glue_app)

return the GlueApplication object?

Need to auto-open the image viewer

The code will look something like

from glue.qt.widgets import ImageWidget
w = glue_app.new_data_viewer(ImageWidget, data=cube_data)

ChrisBeaumont · 2014-10-09T15:07:39Z

spectral_cube/spectral_cube.py

+            result.coords = coordinates_from_header(self.header)
+
+            comp = Component.autotyped(self)
+            result.add_component(comp, name)


I'm pretty sure add_component calls autotyped for you, so you can get rid of the line above and pass self in here

keflavich · 2014-10-09T16:30:53Z

the spectral axis is not WCS, even though the spatial axes are

Can you elaborate on that -- is the problem that the cube is weird, or that Glue isn't displaying something properly, or...?

The slider shows 0...500, but the extracted spectrum is fine, so this is just an issue with the slider. Maybe this is the normal behavior, actually, I don't recall.

there are a lot of warnings and logging messages popping up

I'll take a look. Glue generates lots of logging messages anyways, so it might be benign, and we just need to send them someplace more quietly

Yes, this could be glue interacting with astropy's logging too.

it's not obvious how to send multiple cubes to the same glue app in a nice way - would be best to retrieve the current glue session if possible (the global one as opposed to the SpectralCube._glue_app)

return the GlueApplication object?

This is the problem I described above. For example:

In [2]: gapp = cube303m.to_glue()
Could not parse application stylesheet
UsageError: Invalid GUI request 'inline', valid ones are:['osx', 'qt4', None, 'glut', 'gtk3', 'pyglet', 'wx', 'none', 'qt', 'gtk', 'tk']
In [3]: 2014-10-09 18:22:28.424 python[76840:507] modalSession has been exited prematurely - check for a reentrant call to endModalSession:
/Users/adam/anaconda/envs/astropy27/lib/python2.7/site-packages/matplotlib/text.py:52: UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
  if rotation in ('horizontal', None):
/Users/adam/anaconda/envs/astropy27/lib/python2.7/site-packages/matplotlib/text.py:54: UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
  elif rotation == 'vertical':

gapp
Traceback (most recent call last):
  File "<ipython-input-3-284230671efb>", line 1, in <module>
    gapp
NameError: name 'gapp' is not defined

In this case, Glue has started, but gapp is not even defined as a variable. Note that the above should read:

In [3]: gapp
Traceback (most recent call last):
  File "<ipython-input-3-284230671efb>", line 1, in <module>
    gapp
NameError: name 'gapp' is not defined

but there is other output cluttering the In [3]: prompt.

Need to auto-open the image viewer

The code will look something like
from glue.qt.widgets import ImageWidget
w = glue_app.new_data_viewer(ImageWidget, data=cube_data)

That works. Took me a while to figure out what cube_data was (it is result in this context, i.e. a Dataset)

ChrisBeaumont · 2014-10-09T18:09:44Z

NameError: name 'gapp' is not defined

Ah I think I know whats going on -- see glue-viz/glue#476

This is orthogonal to this PR, so I wouldn't worry about it for now.

ChrisBeaumont · 2014-10-09T18:10:59Z

The slider shows 0...500, but the extracted spectrum is fine, so this is just an issue with the slider.

Yeah that's the intended behavior (slider in pixel coordinates). Maybe it shouldn't be, I'm not sure.

keflavich · 2014-10-09T18:28:35Z

The slider shows 0...500, but the extracted spectrum is fine, so this is just an issue with the slider.

Yeah that's the intended behavior (slider in pixel coordinates). Maybe it shouldn't be, I'm not sure.

It's fine as is, but it would be nice if both could be shown. Especially, for example, a velocity/frequency overlay on the image or titlebar or something of that sort.

keflavich · 2014-10-09T18:34:06Z

So, this works now, but note the slew of warnings:

In [1]: cube = SpectralCube.read('/Users/adam/work/gc/apex/h2co_cubes/APEX_H2CO_303_202.fits')

In [2]: cube2 = SpectralCube.read('/Users/adam/work/gc/apex/h2co_cubes/APEX_H2CO_321_220.fits')

In [3]: gapp = cube.to_glue()
Could not parse application stylesheet
/Users/adam/anaconda/envs/astropy27/lib/python2.7/site-packages/matplotlib/text.py:52: UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
  if rotation in ('horizontal', None):
/Users/adam/anaconda/envs/astropy27/lib/python2.7/site-packages/matplotlib/text.py:54: UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
  elif rotation == 'vertical':
UsageError: Invalid GUI request 'inline', valid ones are:['osx', 'qt4', None, 'glut', 'gtk3', 'pyglet', 'wx', 'none', 'qt', 'gtk', 'tk']
In [4]: cube._glue_app.data_collection[0]
Out[4]: Data (label: SpectralCube)

In [5]: cube2.to_glue(dataset=cube._glue_app.data_collection[0])
INFO:glue.core.hub:Broadcasting DataAddComponentMessage:
     Sent from: Data Set: SpectralCubeNumber of dimensions: 3
Shape: 500 x 200 x 1150
Components:
 0) SpectralCube
 1) Pixel z
 2) Pixel y
 3) Pixel x
 4) Vrad
 5) Galactic Latitude
 6) Galactic Longitude
 7) SpectralCube_
INFO:glue.core.hub:Broadcasting ComponentsChangedMessage:
     Sent from: Data Set: SpectralCubeNumber of dimensions: 3
Shape: 500 x 200 x 1150
Components:
 0) SpectralCube
 1) Pixel z
 2) Pixel y
 3) Pixel x
 4) Vrad
 5) Galactic Latitude
 6) Galactic Longitude
 7) SpectralCube_

Also, @ChrisBeaumont, the GUI is somewhat unresponsive right when I start up - I can't grab the corners to drag/expand the viewing window until I've clicked within the image or activated one of the buttons. Also, I can now reproduce the error I mentioned before, kinda (glue-viz/glue#424): right when the cubes open, they are weirdly downsampled.

keflavich · 2014-10-09T18:34:42Z

@ChrisBeaumont assuming we can clean up some of the error messages etc., what's the next step in terms of writing tests?

ChrisBeaumont · 2014-10-09T18:38:16Z

it would be nice if both could be shown. Especially, for example, a velocity/frequency overlay on the image or titlebar or something of that sort.

The world velocity coordinates are shown along the bottom of the image view when you mouse hover. I've thought about adding a world label to the slider (although it won't have the option to switch between velocity/frequency, since I don't want to be too astro-specific in the UI)

keflavich · 2014-10-09T18:40:00Z

The world velocity coordinates are shown along the bottom of the image view when you mouse hover. I've thought about adding a world label to the slider (although it won't have the option to switch between velocity/frequency, since I don't want to be too astro-specific in the UI)

Ah, OK. And it's fine to not have a switch in the default glue UI; I was thinking the next step for the SpectralCube->Glue viewer is to build a custom scroller with that capability.

ChrisBeaumont · 2014-10-09T18:42:33Z

assuming we can clean up some of the error messages etc., what's the next step in terms of writing tests?

I would add an keyword (at least for testing) that lets you run to_glue without actually opening+running the UI. Then, your tests can just check that the application has the proper datasets, viewers, etc. You can look at GlueApplication.data_collection and .data_viewers to check this.

__getitem__ but that's no problem

the dataset to an existing session

ChrisBeaumont · 2014-11-06T18:49:31Z

glue-viz/glue#501 partially addresses the NameError issue you had before. Glue still hijacks the REPL once you start the UI, but that's only a problem after you stop the glue session, not before you start it. I'm trying to figure out how to address it (it's an IPython thing)

keflavich · 2014-11-14T17:16:41Z

@ChrisBeaumont this has been sitting idle for a while and I don't really have time to dig into it; my side-sprint on the other PR was motivated by immediate need. Do you think this is mergeable or should we just leave it until we can make a reasonable test suite?

ChrisBeaumont · 2014-11-19T21:44:54Z

IMO this is straightforward + functional enough that you could merge it if you want.

Feature request: to_glue

keflavich added help wanted question labels Oct 8, 2014

keflavich reviewed Oct 9, 2014
View reviewed changes

ChrisBeaumont reviewed Oct 9, 2014
View reviewed changes

keflavich added 8 commits October 25, 2014 16:21

first attempt at a to_glue machine. Required fixing some bugs in

e7cfa63

__getitem__ but that's no problem

the glue_app becomes part of the cube now so that subsequent calls add

5c1dfe2

the dataset to an existing session

allow a glue application to be specified. add docstrings

10658e2

attempt to accept updates of data collections

f1e347c

have to add cubes to the dataset object, not the datacollection

948b613

need to check component labels, not 'components'

61fe441

autoload glue viewer

2bd39f1

add start_gui keyword

2613f07

keflavich force-pushed the to_glue branch from c454d55 to 2613f07 Compare October 25, 2014 14:21

keflavich added a commit that referenced this pull request Nov 19, 2014

Merge pull request #153 from keflavich/to_glue

841f208

Feature request: to_glue

keflavich merged commit 841f208 into radio-astro-tools:master Nov 19, 2014

keflavich deleted the to_glue branch November 19, 2014 21:56

keflavich mentioned this pull request Nov 19, 2014

Add tests for glue utility #163

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feature request: to_glue #153

Feature request: to_glue #153

keflavich commented Oct 9, 2014

astrofrog commented Oct 8, 2014

keflavich commented Oct 8, 2014

ChrisBeaumont commented Oct 8, 2014

keflavich commented Oct 9, 2014

keflavich Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont Oct 9, 2014

keflavich Oct 9, 2014

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

keflavich commented Oct 9, 2014

keflavich commented Oct 9, 2014

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont commented Nov 6, 2014

keflavich commented Nov 14, 2014

ChrisBeaumont commented Nov 19, 2014

Feature request: to_glue #153

Feature request: to_glue #153

Conversation

keflavich commented Oct 9, 2014

astrofrog commented Oct 8, 2014

keflavich commented Oct 8, 2014

ChrisBeaumont commented Oct 8, 2014

keflavich commented Oct 9, 2014

keflavich Oct 9, 2014

Choose a reason for hiding this comment

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont Oct 9, 2014

Choose a reason for hiding this comment

keflavich Oct 9, 2014

Choose a reason for hiding this comment

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

keflavich commented Oct 9, 2014

keflavich commented Oct 9, 2014

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

keflavich commented Oct 9, 2014

ChrisBeaumont commented Oct 9, 2014

ChrisBeaumont commented Nov 6, 2014

keflavich commented Nov 14, 2014

ChrisBeaumont commented Nov 19, 2014