apologies if this is a manifest of another issue, though I couldn't find one after some searching.
the following behavior can be seen in 0.14.1 and 0.15.2:
pt = pd.Panel({'a': {'b': {'c': 1, 'd': 1}, 'e': {'c': 2, 'd': 2}}})
pt2 = pd.Panel({'a': {'b': {'c': 3, 'd': 5}, 'e': {'c': 4, 'd': 6}}})
pt.a, pt2.a
mask = (pt.a > 1) & (pt2.a > 4)
pt.a[mask] = pt2.a[mask]
pt.loc[['a'],:,'e']
pt.loc['a',:,'e']
the pt.loc[['a'],:,'e'] shows the original 2 while pt.loc['a',:,'e'] shows the updated 6 from pt2.a
is there a better way to do this, basically i'd like to set the values of pt.a to that of pt2.a based on some mask. any direction would be much appreciated, thanks!
Comment From: jreback
Doing an assignment like pt.a[mask] = value is chained assignment
In [92]: pt.loc['a',:,:] = pt2.a[mask].fillna(pt.a)
In [94]: pt['a',:,'e']
Out[94]:
c 2
d 6
Name: e, dtype: float64
In [95]: pt[['a'],:,'e']
Out[95]:
a
c 2
d 6
So a couple of issues here: - the chained assignment should warn - there are some things that would make this easier
If the mask were constructed to be a Panel itself (this doesn't work ATM, but is easy to fix)
mask = (pt[['a']] > 1) & (pt2.loc[['a']] > 4)
Then
pt[mask] = pt2[mask] also could work
Comment From: ndaniyar
If pt.a[mask] = pt2.a[mask] assignment operates on a copy of pt.a, then how come pt['a',:,'e'] contains the updated values?
Comment From: thinkingatoms
thanks a lot @jreback!
Comment From: jreback
closing as Panels are deprecated