[yt-users] unpickle / load problems

Matthew Turk matthewturk at gmail.com
Fri Mar 15 10:01:21 PDT 2013


Hi Elizabeth,

Okay, I have a fix.  I think.  Unfortunately, pickle files are not
currently tested -- they were, once upon a time, but are not now.
Would you be willing to help out and convert this script to a testing
script?  I think it draws out some important things.

The code I came up with is not terribly pretty.  Unfortunately,
keeping backwards compatibility for pickle files means making a few
compromises in the niceness of the code.  So hopefully this will work,
and not break a bunch of pre-existing pickles.  If it does (paging
David Collins) then it's a bug and we cannot accept the PR.

Here's the pull request:

https://bitbucket.org/yt_analysis/yt/pull-request/463/attempting-to-create-a-backwards

Let me know if it works for you.  The changeset hash to pull:

hg pull -r 15d72caf0ea6 https://bitbucket.org/MatthewTurk/yt/

-Matt

On Fri, Mar 15, 2013 at 11:59 AM, Matthew Turk <matthewturk at gmail.com> wrote:
> Hi Elizabeth,
>
> Okay -- cool.  I think I will work around the h5py thing for now.  But
> it is worrisome; I think save_object is currently broken as h5py
> changed how it deals with VLEN strings between 2.0 and 2.1.
>
> I've reproduced it by using Pickle by itself.  More as I get it.
>
> -Matt
>
> On Fri, Mar 15, 2013 at 11:25 AM, Elizabeth Tasker
> <tasker at astro1.sci.hokudai.ac.jp> wrote:
>> Hi Matt,
>>
>> According to the directory I have in yt, an older version:
>>
>> /Users/Elizabeth/yt/src/h5py-2.0.1/
>>
>> Elizabeth
>>
>> On Mar 15, 2013, at 10:04 AM, Matthew Turk <matthewturk at gmail.com> wrote:
>>
>>> Hi Elizabeth,
>>>
>>> On Wed, Mar 13, 2013 at 4:45 PM, Elizabeth Tasker
>>> <tasker at astro1.sci.hokudai.ac.jp> wrote:
>>>> Hi Matt,
>>>>
>>>> So ironically, it took me a long time to find the right combination to make this break in the test script! The fatal combination seems to be an extract region from a boolean object.
>>>>
>>>> This is the script I used:
>>>>
>>>> http://paste.yt-project.org/show/3228/
>>>>
>>>> It should work with any data set.
>>>>
>>>> It creates 6 saved objects. Objects 1-4 work fine but 5 and 6 have the attribute error.
>>>
>>> I'm getting h5py errors when I even run the script.  Which version of
>>> h5py are you on?
>>>
>>>>>> h5py.version.version
>>> '2.1.0'
>>>
>>>
>>>>
>>>> Elizabeth
>>>>
>>>>
>>>> On Mar 13, 2013, at 3:08 PM, Matthew Turk <matthewturk at gmail.com> wrote:
>>>>
>>>>> Hi Elizabeth,
>>>>>
>>>>> Okay, this seems like it's bumped up against a problem with the way yt
>>>>> handles pickles.  At some point in the reconstruction, data objects
>>>>> aren't getting separated from their parameter files.
>>>>>
>>>>> Any chance you could make something that reproduces this on one of the
>>>>> workshop datasets?
>>>>>
>>>>> http://yt-project.org/data/
>>>>>
>>>>> -Matt
>>>>>
>>>>> On Wed, Mar 13, 2013 at 3:03 PM, Elizabeth Tasker
>>>>> <tasker at astro1.sci.hokudai.ac.jp> wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I'm struggling to unpickle / load_object a previously pickled / save_object.
>>>>>>
>>>>>> I've done:
>>>>>>
>>>>>> pf.h.save_object(allclouds, "clouds_pyfindclouds2_yt")
>>>>>>
>>>>>> but when I do:
>>>>>>
>>>>>> clouds = pf.h.load_object("clouds_pyfindclouds2_yt")
>>>>>>
>>>>>> I get:
>>>>>>
>>>>>> AttributeError                            Traceback (most recent call last)
>>>>>> /Users/Elizabeth/yt/src/yt-hg/scripts/iyt in <module>()
>>>>>> ----> 1 execfile("cloud_surfacedensity_b.py")
>>>>>>
>>>>>> /Users/Elizabeth/yt/src/yt-hg/scripts/iyt in <module>()
>>>>>>    13 #file.close()
>>>>>>    14
>>>>>> ---> 15 clouds = pf.h.load_object("clouds_pyfindclouds2_yt")
>>>>>>
>>>>>> /Users/Elizabeth/yt/src/yt-hg/yt/data_objects/hierarchy.pyc in load_object(self, name)
>>>>>>   328         if obj is None:
>>>>>>   329             return
>>>>>> --> 330         obj = cPickle.loads(obj.value)
>>>>>>   331         if iterable(obj) and len(obj) == 2:
>>>>>>   332             obj = obj[1] # Just the object, not the pf
>>>>>>
>>>>>> /Users/Elizabeth/yt/src/yt-hg/yt/data_objects/data_containers.pyc in _reconstruct_object(*args, **kwargs)
>>>>>>  4623     pf = pfs.get_pf_hash(pfid)
>>>>>>  4624     cls = getattr(pf.h, dtype)
>>>>>> -> 4625     obj = cls(*new_args)
>>>>>>  4626     obj.field_parameters.update(field_parameters)
>>>>>>  4627     return pf, obj
>>>>>>
>>>>>> /Users/Elizabeth/yt/src/yt-hg/yt/data_objects/data_containers.pyc in __init__(self, regions, fields, pf, **kwargs)
>>>>>>  4046         self._all_overlap = []
>>>>>>  4047         self._cut_masks = {}
>>>>>> -> 4048         self._get_all_regions()
>>>>>>  4049         self._make_overlaps()
>>>>>>  4050         self._get_list_of_grids()
>>>>>>
>>>>>> /Users/Elizabeth/yt/src/yt-hg/yt/data_objects/data_containers.pyc in _get_all_regions(self)
>>>>>>  4057             self._all_regions.append(item)
>>>>>>  4058             # So cut_masks don't get messed up.
>>>>>> -> 4059             item._boolean_touched = True
>>>>>>  4060         self._all_regions = np.unique(self._all_regions)
>>>>>>  4061
>>>>>>
>>>>>> AttributeError: ("'tuple' object has no attribute '_boolean_touched'", <function _reconstruct_object at 0x101dcd578>, ('73850deee907adf5217f43156c5057cd', 'boolean', [(GR_Enzo2_128amr5_zeus_turb_psupp_rr_0240, AMRCylinder (GR_Enzo2_128amr5_zeus_turb_psupp_rr_0240): center=[ 16.  16.  16.], _norm_vec=[ 0.  0.  1.], _radius=7.5, _height=0.100000208264), 'NOT', (GR_Enzo2_128amr5_zeus_turb_psupp_rr_0240, AMRCylinder (GR_Enzo2_128amr5_zeus_turb_psupp_rr_0240): center=[ 16.  16.  16.], _norm_vec=[ 0.  0.  1.], _radius=4.5, _height=0.100000208264)], {'bulk_velocity': array([ 0.,  0.,  0.]), 'center': array([ 0.5,  0.5,  0.5]), 'normal': array([ 0.,  0.,  1.])}))
>>>>>>
>>>>>>
>>>>>>
>>>>>> This is all on the same computer.
>>>>>>
>>>>>> Elizabeth
>>>>>> _______________________________________________
>>>>>> yt-users mailing list
>>>>>> yt-users at lists.spacepope.org
>>>>>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>>>> _______________________________________________
>>>>> yt-users mailing list
>>>>> yt-users at lists.spacepope.org
>>>>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>>>
>>>> _______________________________________________
>>>> yt-users mailing list
>>>> yt-users at lists.spacepope.org
>>>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>> _______________________________________________
>>> yt-users mailing list
>>> yt-users at lists.spacepope.org
>>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>
>> _______________________________________________
>> yt-users mailing list
>> yt-users at lists.spacepope.org
>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org



More information about the yt-users mailing list