add weighted ppc #2273
add weighted ppc #2273
Conversation
pymc3/sampling.py
Outdated
| @@ -536,6 +544,100 @@ def sample_ppc(trace, samples=None, model=None, vars=None, size=None, | |||
| return {k: np.asarray(v) for k, v in ppc.items()} | |||
|
|
|||
|
|
|||
| def sample_ppc_w(traces, samples=None, models=None, size=None, weights=None, | |||
There was a problem hiding this comment.
IMO Your approach before was better - why adding a new function instead?
There was a problem hiding this comment.
I think it could be potentially confusing for the user.
|
I think is better for the user to have a separated function to perform a weighted ppc (model averaging). |
|
I see, a notebook demonstrating how to use this for model averaging is necessary then. It would be quite a cool feature. |
|
I will try do add that notebook ASAP, thanks for the review! |
pymc3/sampling.py
Outdated
| raise ValueError( | ||
| 'The number of traces and weights should be the same') | ||
| indices = randint(0, len(trace), samples) | ||
| if progressbar: |
There was a problem hiding this comment.
tqdm progress bars need to be closed if sampling is interrupted. Some try...finally would do it.
|
I add an example showing how to use the new function and fix the tqdm issue |
|
Nice! I think it's ready to merge once the test pass. |
|
This looks nice! But I think we should merge only after the 3.1 release. |
|
Fine with me. |
|
@junpenglao After we fix the notebooks and examples we should be good to go. |
fix error docstring
|
I just fix a couple of conflicts, that I think were introduced by #2291 |
aloctavodia
force-pushed
the
sample_ppc_ma
branch
from
db0da22 to
e282df2
Compare
|
I just fix a couple of conflicts (probably in the less elegant way). |
|
Thanks @aloctavodia |
Allows to use
sample_ppcto get weighted samples, if more than one trace and a set of weights are passed.The weights could come from the
comparefunction, for example.I will try to add a notebook with examples ASAP.