How Shape Information is Handled by Theano?
This recipe explains how shape information is handled by theano.
Recipe Objective - How Shape Information is Handled by Theano?
In Theano, the shape information is currently employed in two ways:
1. When the precise output shape is known in advance, develop quicker C code for the 2d convolution on the CPU and GPU.
2. To eliminate computations from the graph when we simply want to know the shape of a variable rather than its actual value. The Op.infer shape method is used to accomplish this.
Access Face Recognition Project Code using Facenet in Python
For more related projects -
/projects/data-science-projects/tensorflow-projects
/projects/data-science-projects/neural-network-projects
Example -
# Import libraries
import theano
from theano import tensor
# Creating a matrix
a = tensor.dmatrix('a')
# Creaing a function
fun = theano.function([a], (a ** 3).shape)
# Debug
theano.printing.debugprint(fun)
Output -
MakeVector{dtype='int64'} [id A] '' 2
|Shape_i{0} [id B] '' 1
| |a [id C]
|Shape_i{1} [id D] '' 0
|a [id C]
The output of this compiled function directly shows the shape of the output, instead of any multiplication or power.
What Users are saying..
Abhinav Agarwal
I come from Northwestern University, which is ranked 9th in the US. Although the high-quality academics at school taught me all the basics I needed, obtaining practical experience was a challenge.... Read More