Text backend does not support tallying a subset of variables (raises KeyError)

[chumsley]

Attempting to use a trace that was loaded from the text backend yields a KeyError, if the trace contains only a subset of the model's variables (e.g., if it was created using the trace= argument to sample).

Failing example:

with pm.Model() as model:
    a = pm.Gamma('a', mu=10.0, sd=2.0)
    b = pm.Gamma('b', mu=10.0, sd=2.0)

    trace = pm.sample(trace=[model.a, model.a_log__])
    assert len(trace.varnames) == 2

    pm.backends.text.dump('trace.text', trace)

    loaded = pm.backends.text.load('trace.text')
    x = loaded[0] #!!! Will throw a KeyError looking for 'b_log__'

[junpenglao]

Thanks for reporting. We should add the vars=None kwarg into pm.backends.text.load and pm.backends.text.dump

[haritha1313]

@twiecki @junpenglao I'd like to work on this issue.

[vermashresth]

Hi @junpenglao
I have gone through the code base of pymc3 and would like to start working on an issue. This one seems old but is it still relevant?

[twiecki]

@vermashresth Definitely, you can just try the above code to see if the problem persists.

[michaelosthege]

We're deprecating multiple backends.. (see #3903)
While the text backend may survive a bit longer, only very few people seem to be using it.

As indicated by our deprecation warnings, please switch backends. If you can't please contact us and tell us why.

Closing this issue, because most likely, nobody will take it up.