2

In certain cases pymc3 automatically performs variable transformations (see What are the interval transforms in pymc3 for uniform distributions?, what does the 'find_MAP' output mean in pymc3?).

In answer https://stackoverflow.com/a/42170938/5142797 I found which transformations are applied. So in principal I can perform backward and forward transformations now.

But is there a more convenient way to perform these forward and backward transformations on automatically transformed variables? That would be very convenient, e.g. for find_MAP results.

Community
  • 1
  • 1
Fabian Rost
  • 2,284
  • 2
  • 15
  • 27

1 Answers1

3

The following works:

test_model = pymc3.Model()

with test_model:
    u = pymc3.Uniform('u', 0, 1)

test_model.u_interval__.distribution.transform_used.backward(0.5).eval()
Fabian Rost
  • 2,284
  • 2
  • 15
  • 27