In [1]:

    

import theano.tensor as T
x = T.dmatrix("x")
y = T.dmatrix("y")
z = x + y


    

In [2]:

    

type(z)


    

    
Out[2]:





theano.tensor.var.TensorVariable

In [5]:

    

type(z.owner)


    

    
Out[5]:





theano.gof.graph.Apply

In [7]:

    

z.owner.op.name


    

    
Out[7]:





'Elemwise{add,no_inplace}'

In [8]:

    

w = x * 2
w.owner.op.name


    

    
Out[8]:





'Elemwise{mul,no_inplace}'

In [13]:

    

print(w.owner)
print(len(w.owner.inputs))
print(w.owner.inputs) # second input is not 2


    

    




Elemwise{mul,no_inplace}(x, DimShuffle{x,x}.0)
2
[x, DimShuffle{x,x}.0]

In [14]:

    

print(z.owner)
print(len(z.owner.inputs))
print(z.owner.inputs)


    

    




Elemwise{add,no_inplace}(x, y)
2
[x, y]

In [18]:

    

w.owner.inputs[1].owner.inputs


    

    
Out[18]:





[TensorConstant{2}]

In [1]:

    

# graph visualization
import theano
import pydot
v = theano.tensor.vector()
from IPython.display import SVG
SVG(theano.printing.pydotprint(v*2, return_image=True,
                               format='svg'))


    

    
Out[1]:

In [2]:

    

# optimization visualization
import theano
import pydot
a = theano.tensor.vector("a")
b = a + a ** 10
f = theano.function([a], b)
print(f([0,1,2]))


    

    




[    0.     2.  1026.]

In [6]:

    

theano.printing.pydotprint(b, outfile = './grap.png', var_with_name_simple=True)


    

    




The output file is available at ./grap.png

In [8]:

    

theano.printing.pydotprint(f, outfile="./graph_opt.png", var_with_name_simple=True)


    

    




The output file is available at ./graph_opt.png

In [ ]: