BUG: `Trajectories()` objects don't get saved transparently in a HDFStore, raises TypeError

[glyg]

With trajs being a Trajectories object, this fails:

import pandas as pd
with pd.get_store('with_gaps.h5') as store:
    store['trajs'] = trajs

raises:

TypeError: cannot properly create the storer for:
 [_TYPE_MAP] [group->/trajs (Group) '',
value-><class'sktracker.trajectories.trajectories.Trajectories'>,
format->fixed,
append->False,kwargs->{'encoding': None}]

I think it's a bug, because users will assume they can do that. Of course, the following works smoothly:

import pandas as pd
with pd.get_store('with_gaps.h5') as store:
    store['trajs'] = pd.DataFrame(trajs)

I'll post something in pandas to see if there's a way around that on their side (after all, sub classing should be supported). I'm not sure what to do on our side... Of course, objectsio should take care of that, but I'd feel better if that kind of TypeError won't appear at all

[glyg]

Might be related:

http://stackoverflow.com/questions/22155951/how-to-subclass-pandas-dataframe

[glyg]

Much more related:
pandas-dev/pandas#4271

On the thread, you can find this:

import pandas as pd
from pandas.io import pytables 

class MyPanel(pd.Panel):
    pass

class MyPanelStorer(pytables.PanelStorer):
    pandas_kind = 'mywide'
    obj_type = MyPanel
pytables._TYPE_MAP[MyPanel] = 'mywide'
pytables._STORER_MAP['mywide'] = 'MyPanelStorer'
pytables.MyPanelStorer = MyPanelStorer 

Turns out this is outdated a bit, but I think the best solution is to add that:

pytables._TYPE_MAP[Trajectories] = 'frame'

Which basically writes trajectories as DataFrames. The simpler the better I guess

[hadim]

I agree that we should support that. People should be able to run tracking and save results with Trajectories without using ObjectsIO at all.