game_theory\support_enumeration.py

[guihigashi]

I want to ask if this works on some other OS (I use windows). This code causes numba to complain about the zip in the for loop at line 104, and the same function works if i remove the argument nopython=True, default is false. Maybe the problem is my system or should i make a PR?
Thanks

[jstac]

If you type

import numba
numba.__version__

what do you get?

[guihigashi]

0.39.0

[jstac]

@guihigashi Thanks for reporting this. Would you mind to copy and paste a minimal piece of code that generates the error, along with the error message? We can then test it on a few machines.

[guihigashi]

this is from the game theory notebook

import numpy as np
import quantecon.game_theory as gt

matching_pennies_bimatrix = [[(1, -1), (-1, 1)],
                             [(-1, 1), (1, -1)]]

g_MP = gt.NormalFormGame(matching_pennies_bimatrix)

gt.support_enumeration(g_MP) # error happens here

Error message

---------------------------------------------------------------------------
TypingError                               Traceback (most recent call last)
<ipython-input-52-deeb4068f149> in <module>()
----> 1 gt.support_enumeration(g_MP)

~\Miniconda3\envs\py36\lib\site-packages\quantecon\game_theory\support_enumeration.py in support_enumeration(g)
     37 
     38     """
---> 39     return list(support_enumeration_gen(g))
     40 
     41 

~\Miniconda3\envs\py36\lib\site-packages\quantecon\game_theory\support_enumeration.py in support_enumeration_gen(g)
     61     if N != 2:
     62         raise NotImplementedError('Implemented only for 2-player games')
---> 63     return _support_enumeration_gen(g.payoff_arrays)
     64 
     65 

~\Miniconda3\envs\py36\lib\site-packages\numba\dispatcher.py in _compile_for_args(self, *args, **kws)
    347                 e.patch_message(msg)
    348 
--> 349             error_rewrite(e, 'typing')
    350         except errors.UnsupportedError as e:
    351             # Something unsupported is present in the user code, add help info

~\Miniconda3\envs\py36\lib\site-packages\numba\dispatcher.py in error_rewrite(e, issue_type)
    314                 raise e
    315             else:
--> 316                 reraise(type(e), e, None)
    317 
    318         argtypes = []

~\Miniconda3\envs\py36\lib\site-packages\numba\six.py in reraise(tp, value, tb)
    656             value = tp()
    657         if value.__traceback__ is not tb:
--> 658             raise value.with_traceback(tb)
    659         raise value
    660 

TypingError: Failed at nopython (nopython frontend)
Invalid usage of Function(<class 'zip'>) with parameters ((array(int64, 1d, C), array(int32, 1d, C)), tuple(array(float64, 1d, C) x 2))
 * parameterized
In definition 0:
    All templates rejected
[1] During: resolving callee type: Function(<class 'zip'>)
[2] During: typing of call at C:\Users\guihi\Miniconda3\envs\py36\lib\site-packages\quantecon\game_theory\support_enumeration.py (104)


File "..\..\..\..\Miniconda3\envs\py36\lib\site-packages\quantecon\game_theory\support_enumeration.py", line 104:
def _support_enumeration_gen(payoff_matrices):
    <source elided>
                        for p, (supp, action) in enumerate(zip(supps,
                                                               actions)):
                                                               ^

[jstac]

This works fine for me on Linux, numba version 0.39.

@mmcky @oyamad Do you a Windows machine you can test on?

[oyamad]

No, I don't have access to a Windows machine myself. I will try to ask some of my students later today.

@guihigashi Are you on a 32-bit machine?

[guihigashi]

64-bit

[oyamad]

@guihigashi Can you show the out put of the following?

import numpy as np
np.int_

[guihigashi]

<class 'numpy.int32'>

[guihigashi]

ok, it works if i change line 88 from supps = (np.arange(k), np.empty(k, np.int_)) to supps = (np.arange(k), np.empty(k, np.int64))

[oyamad]

Thanks.

A workaround should be the following:
Replace

QuantEcon.py/quantecon/game_theory/support_enumeration.py

Line 88 in 64918ce

supps = (np.arange(k), np.empty(k, np.int_))

with

        supps = (np.arange(0, k, 1, np.int_), np.empty(k, np.int_))

I will make a PR later today.

[oyamad]

@guihigashi Good, thanks. But I think it would be better to guarantee the dtypes of the elements of the supps tuple to be the same (on any platform).

[guihigashi]

thanks, I'll stick with your solution.
was it because the tuple was mixing 32 bits with 64 bits?

[guihigashi]

yes, of course. Code portability.

[oyamad]

Yes, I believe that's the cause of the error.