Cannot create a Normal distribution with carl 0.2, Theano 0.8.2

[gnperdue]

Hello,

I am attempting to test this library, but running into problems quickly. The versions of relevant libraries I have are:

$ python pyverchecker.py
Python version: 2.7.13 |Continuum Analytics, Inc.| (default, Dec 20 2016, 23:05:08)
[GCC 4.2.1 Compatible Apple LLVM 6.0 (clang-600.0.57)]
pandas version: 0.19.2
matplotlib version: 2.0.0
numpy version: 1.12.0
scipy version: 0.19.0
IPython version: 5.3.0
sklearn version: 0.18
pymc3 version: 3.0
theano version: 0.8.2
carl version: 0.2
h5py version: 2.6.0

If I try to instantiate a Normal, I get:

(py2th) ~$ ipython
Python 2.7.13 |Continuum Analytics, Inc.| (default, Dec 20 2016, 23:05:08)
Type "copyright", "credits" or "license" for more information.

IPython 5.3.0 -- An enhanced Interactive Python.
?         -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help      -> Python's own help system.
object?   -> Details about 'object', use 'object??' for extra details.

In [1]: from carl.distributions import Normal

In [2]: pnorm = Normal(mu=0.0, sigma=1.0)
ERROR (theano.gof.opt): Optimization failure due to: local_mul_canonizer
ERROR (theano.gof.opt): node: Elemwise{true_div,no_inplace}(Elemwise{mul,no_inplace}.0, Elemwise{pow,no_inplace}.0)
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/Users/gnperdue/anaconda/envs/py2th/lib/python2.7/site-packages/theano/gof/opt.py", line 1772, in process_node
    replacements = lopt.transform(node)
  File "/Users/gnperdue/anaconda/envs/py2th/lib/python2.7/site-packages/theano/tensor/opt.py", line 4505, in transform
    new = self.merge_num_denum(num, denum)
  File "/Users/gnperdue/anaconda/envs/py2th/lib/python2.7/site-packages/theano/tensor/opt.py", line 4327, in merge_num_denum
    return self.inverse(self.merge_num_denum(num, []),
  File "/Users/gnperdue/anaconda/envs/py2th/lib/python2.7/site-packages/theano/tensor/opt.py", line 4326, in merge_num_denum
    return self.main(*num)
  File "/Users/gnperdue/anaconda/envs/py2th/lib/python2.7/site-packages/theano/gof/op.py", line 633, in __call__
    (i, ins, node))
ValueError: Cannot compute test value: input 1 (Elemwise{pow,no_inplace}.0) of Op Elemwise{mul,no_inplace}(DimShuffle{x,x}.0, Elemwise{pow,no_inplace}.0) missing default value

There is actually a long spew of errors, all ending in ValueErrors that complain about "missing default values."

Any suggestions?

Thanks!

pax
Gabe

[glouppe]

Thanks for the report! I'll try looking into this asap.

(I checked on a fresh Python 3 environment and all works fine. Certainly due to Python 2.)

[gnperdue]

Thanks. I forgot to mention I am using a Mac and installed basically everything via conda-forge.

[gnperdue]

I can confirm this works for me with Python 3 (but different Theano version - 0.9) - so this does look like a Python 2 issue:

(py3th) $ python pyverchecker.py
Python version: 3.5.3 |Continuum Analytics, Inc.| (default, Mar  6 2017, 12:15:08)
[GCC 4.2.1 Compatible Apple LLVM 6.0 (clang-600.0.57)]
pandas version: 0.19.2
matplotlib version: 2.0.0
numpy version: 1.12.1
scipy version: 0.19.0
IPython version: 5.3.0
sklearn version: 0.18.1
tensorflow version: 1.0.0
pymc3 version: 3.0
theano version: 0.9.0
carl version: 0.2
h5py version: 2.7.0
Using TensorFlow backend.
keras version: 2.0.2
Note, Keras configuration is in $HOME/.keras/keras.json
(py3th) $ ipython
Python 3.5.3 |Continuum Analytics, Inc.| (default, Mar  6 2017, 12:15:08)
Type "copyright", "credits" or "license" for more information.

IPython 5.3.0 -- An enhanced Interactive Python.
?         -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help      -> Python's own help system.
object?   -> Details about 'object', use 'object??' for extra details.

In [1]: from carl.distributions import Normal

In [2]: pnorm = Normal(mu=0.0, sigma=1.0)

In [3]:

I posted my Anaconda config script here (should it somehow matter):

https://github.com/gnperdue/BashDotFiles/tree/master/anaconda_configs

After setting up Anaconda, I just clone and install Carl via setup.py.