python - How should I allocate a numpy array inside theano function? -


let's have theano function:

def my_fun(x, y):   # create output array example sake   z = np.asarray(     shape=(x.shape[0], y.shape[1]),     dtype=theano.config.floatx   )    z = x + y    # wrong, how should convert theano   # tensor?   return z  x = theano.tensor.dmatrix("x") y = theano.tensor.dmatrix("y")  f = function(   inputs=[x, y],   outputs=[my_fun] )  = numpy.asarray([[1,2],[3,4]]) b = numpy.asarray([[1,2],[3,4]])  c = my_fun(a,b) 
  1. how should allocate tensors/ arrays or memory within actual theano optimized when compiled theano.
  2. how should convert allocated tensor/ array whatever theano variable returned? i've tried converting shared variable in function didn't work.

i'm sorry don't understand specific questions can comment on code sample provided.

firstly, comment above return z incorrect. if x , y theano variables z theano variable after z = x + y.

secondly, there no need pre-allocate memory, using numpy, return variables. my_fun can change simply

def my_fun(x, y):   z = x + y   return z 

thirdly, output(s) of theano functions need theano variables, not python functions. , output needs function of inputs. theano.function call needs changed to

f = function(   inputs=[x, y],   outputs=[my_fun(x, y)] ) 

the important point grasp theano, can little difficult one's head around when starting out, difference between symbolic world , executable world. tied in difference between python expressions , theano expressions.

the modified my_fun above used symbolic function or normal executable python function behaves differently each. if pass in normal python inputs addition operation occurs , return value result of computation. my_fun(1,2) returns 3. if instead pass in symbolic theano variables addition operation not take place immediately. instead function returns symbolic expression after later being compiled , executed return result of adding 2 inputs. result of my_fun(theano.tensor.scalar(), theano.tensor.scalar()) python object represents symbolic theano computation graph. when result passed output theano.function compiled executable. thean when compiled function executed, , given concrete values inputs, result looking for.


Comments

Popular posts from this blog

python - Healpy: From Data to Healpix map -

c - Bitwise operation with (signed) enum value -

xslt - Unnest parent nodes by child node -