sympy to theano conversion failure for complex expressions

[Amir Farbin]

Hi,

I'm trying to convert expressions sympy to theano... following:

http://matthewrocklin.com/blog/work/2013/03/28/SymPy-Theano-part-2/

And I run into a failure that appears simple to fix. Here's a simple example:

from sympy.printing.theanocode import theano_function

im=sp.sqrt(-1)

x=sp.symbols("x")

y=x+im*x

fn_theano  = theano_function([x], [y], dims={x: 1}, dtypes={x: 'float64'})

ends with:

KeyError: <class 'sympy.core.numbers.ImaginaryUnit'>

which appears to my naive eyes to be a just a missing entry in a conversion map. Can someone help? 

Thanks,

Amir

[Ondřej Čertík]

On Mon, Nov 10, 2014 at 10:54 AM, Amir Farbin <afa...@gmail.com> wrote:
> Hi,
>
>
> I'm trying to convert expressions sympy to theano... following:
>
>
> http://matthewrocklin.com/blog/work/2013/03/28/SymPy-Theano-part-2/
>
>
> And I run into a failure that appears simple to fix. Here's a simple
> example:
>
>
> from sympy.printing.theanocode import theano_function
>
>
> im=sp.sqrt(-1)

Note that you can use "I", i.e. sp.I.

>
> x=sp.symbols("x")
>
> y=x+im*x
>
> fn_theano = theano_function([x], [y], dims={x: 1}, dtypes={x: 'float64'})
>
>
> ends with:
>
>
> KeyError: <class 'sympy.core.numbers.ImaginaryUnit'>
>
>
> which appears to my naive eyes to be a just a missing entry in a conversion
> map. Can someone help?

There seem to be more bugs. With the latest master, I am getting:


In [1]: from sympy.printing.theanocode import theano_function

In [3]: import sympy as sp

In [4]: im=sp.sqrt(-1)

In [5]: x=sp.symbols("x")

In [6]: y=x+im*x

In [7]: fn_theano = theano_function([x], [y], dims={x: 1}, dtypes={x:
'float64'})
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
<ipython-input-7-8d3dbb9c015b> in <module>()
----> 1 fn_theano = theano_function([x], [y], dims={x: 1}, dtypes={x:
'float64'})

/home/certik/repos/sympy/sympy/printing/theanocode.py in
theano_function(inputs, outputs, dtypes, cache, **kwargs)
224 code = partial(theano_code, cache=cache, dtypes=dtypes,
225 broadcastables=broadcastables)
--> 226 tinputs = list(map(code, inputs))
227 toutputs = list(map(code, outputs))
228 toutputs = toutputs[0] if len(toutputs) == 1 else toutputs

/home/certik/repos/sympy/sympy/printing/theanocode.py in
theano_code(expr, cache, **kwargs)
192
193 def theano_code(expr, cache=global_cache, **kwargs):
--> 194 return TheanoPrinter(cache=cache,
settings={}).doprint(expr, **kwargs)
195
196

/home/certik/repos/sympy/sympy/printing/theanocode.py in doprint(self,
expr, **kwargs)
187 def doprint(self, expr, **kwargs):
188 """Returns printer's representation for expr (as a string)"""
--> 189 return self._print(expr, **kwargs)
190
191 global_cache = {}

/home/certik/repos/sympy/sympy/printing/printer.pyc in _print(self,
expr, *args, **kwargs)
255 printmethod = '_print_' + cls.__name__
256 if hasattr(self, printmethod):
--> 257 return getattr(self, printmethod)(expr,
*args, **kwargs)
258
259 # Unknown object, fall back to the emptyPrinter.

/home/certik/repos/sympy/sympy/printing/theanocode.py in
_print_Symbol(self, s, dtypes, broadcastables)
78 return self.cache[key]
79 else:
---> 80 value = tt.tensor(name=s.name, dtype=dtype,
broadcastable=broadcastable)
81 self.cache[key] = value
82 return value

NameError: global name 'tt' is not defined




Thanks for letting us know. This might be quite easy to fix, just go
into the source file, do the fix and test it. We can help you out. If
you fix it, then just send us a pull request.

Ondrej

[Ondřej Čertík]

Ah, ok, this is caused by me not having the theano module installed,
i.e. from the beginning of sympy/printing/theanocode.py:

theano = import_module('theano')
if theano:
ts = theano.scalar
tt = theano.tensor

otherwise 'tt' is not defined.

Anyway, Amir, try to add ImaginaryUnit into the "mapping" dicitonary
in sympy/printing/theanocode.py, that should do it. Send us a PR with
the fix.

Ondrej

[Amir Farbin]

Hi,

I couldn't find the theano representation of imaginary unit (that's why asked actually). Anyway, the following addition the the mapping worked for me:

sympy.numbers.ImaginaryUnit: lambda:tt.sqrt(-1),

Thanks,

Amir

[Amir Farbin]

Correction: I guess this is the right thing to do:

sympy.numbers.ImaginaryUnit: lambda:tt.complex(0,1),

Sorry for the mistake.

Amir

[Frédéric Bastien]

That seem good.

It would be great if you can make a PR to sympy to add this.

Fred

--
You received this message because you are subscribed to the Google Groups "sympy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
To post to this group, send email to sy...@googlegroups.com.
Visit this group at http://groups.google.com/group/sympy.

To view this discussion on the web visit https://groups.google.com/d/msgid/sympy/ee9e900a-c1d7-4905-b80d-1319f3dc1350%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.