python - Why am I allowed pickle instancemethods that are Theano functions, but not normal instancemethods? -

in process of using joblib parallelize model-fitting code involving theano functions, i've stumbled across behavior seems odd me.

consider simplified example:

from joblib import parallel, delayed import theano theano import tensor te import numpy np  class theanomodel(object):     def __init__(self):         x = te.dvector('x')         y = (x ** te.log(x ** 2)).sum()         self.theano_get_y = theano.function([x], y)      def get_y(self, x):         return self.theano_get_y(x)  def run(niter=100):     x = np.random.randn(1000)     model = theanomodel()     pool = parallel(n_jobs=-1, verbose=1, pre_dispatch='all')      # fails `typeerror: can't pickle instancemethod objects`...     results = pool(delayed(model.get_y)(x) _ in xrange(niter))      # # ... works! why?     # results = pool(delayed(model.theano_get_y)(x) _ in xrange(niter))  if __name__ == '__main__':     run() 

i understand why first case fails, since .get_y() instancemethod of theanomodel. don't understand why second case works, since x, y andtheano_get_y() declared within __init__() method of theanomodel. theano_get_y() can't evaluated until theanomodel instance has been created. surely, then, should considered instancemethod, , should therefore unpickleable? in fact, still works if explicitly declare x , y attributes of theanomodel instance.

can explain what's going on here?

---

update

just illustrate why think behaviour particularly weird, here few examples of other callable member objects don't take self first argument:

from joblib import parallel, delayed import theano theano import tensor te import numpy np  class theanomodel(object):     def __init__(self):         x = te.dvector('x')         y = (x ** te.log(x ** 2)).sum()         self.theano_get_y = theano.function([x], y)         def square(x):             return x ** 2         self.member_function = square         self.static_method = staticmethod(square)         self.lambda_function = lambda x: x ** 2  def run(niter=100):     x = np.random.randn(1000)     model = theanomodel()     pool = parallel(n_jobs=-1, verbose=1, pre_dispatch='all')      # # not allowed: `typeerror: can't pickle function objects`     # results = pool(delayed(model.member_function)(x) _ in xrange(niter))      # # not allowed: `typeerror: can't pickle function objects`     # results = pool(delayed(model.lambda_function)(x) _ in xrange(niter))      # # not allowed: `typeerror: can't pickle staticmethod objects`     # results = pool(delayed(model.static_method)(x) _ in xrange(niter))      # totally fine!?     results = pool(delayed(model.theano_get_y)(x) _ in xrange(niter))  if __name__ == '__main__':     run() 

none of them pickleable exception of theano.function!

theano functions aren't python functions. instead python objects override __call__. means can call them function internally objects of custom class. in consequence, can pickle them.