Sparse Tensors



In [1]:

    
from __future__ import print_function
import tensorflow as tf
import numpy as np



In [2]:

    
from datetime import date
date.today()









    Out[2]:





datetime.date(2017, 7, 19)



In [3]:

    
author = "kyubyong. https://github.com/Kyubyong/tensorflow-exercises"



In [4]:

    
tf.__version__









    Out[4]:





'1.2.1'



In [5]:

    
np.__version__









    Out[5]:





'1.13.1'



In [6]:

    
sess = tf.InteractiveSession()

Sparse Tensor Representation & Conversion

Q1. Convert tensor x into a SparseTensor.



In [61]:

    
x = tf.constant([[1, 0, 0, 0],
                 [0, 0, 2, 0],
                 [0, 0, 0, 0]], dtype=tf.int32)
sp = ...
print(sp.eval())









    



SparseTensorValue(indices=array([[0, 0],
       [1, 2]]), values=array([1, 2], dtype=int32), dense_shape=array([3, 4]))

Q2. Investigate the dtype, indices, dense_shape and values of the SparseTensor sp in Q1.



In [62]:

    
print("dtype:", ...)
print("indices:", ...)
print("dense_shape:", ...)
print("values:", ...)









    



dtype: <dtype: 'int32'>
indices: [[0 0]
 [1 2]]
dense_shape: [3 4]
values: [1 2]

Q3. Let's write a custom function that converts a SparseTensor to Tensor. Complete it.



In [63]:

    
def dense_to_sparse(tensor):
    indices = tf.where(tf.not_equal(tensor, 0))
    return tf.SparseTensor(indices=indices,
                           values=...,  # for zero-based index
                           dense_shape=tf.to_int64(tf.shape(tensor)))

# Test
print(dense_to_sparse(x).eval())









    



SparseTensorValue(indices=array([[0, 0],
       [1, 2]]), values=array([0, 1], dtype=int32), dense_shape=array([3, 4]))

Q4. Convert the SparseTensor sp to a Tensor using tf.sparse_to_dense.



In [65]:

    
output = ...
print(output.eval())
print("Check if this is identical with x:\n", x.eval())









    



[[1 0 0 0]
 [0 0 2 0]
 [0 0 0 0]]
Check if this is identical with x:
 [[1 0 0 0]
 [0 0 2 0]
 [0 0 0 0]]

Q5. Convert the SparseTensor sp to a Tensor using tf.sparse_tensor_to_dense.



In [67]:

    
output = ...
print(output.eval())
print("Check if this is identical with x:\n", x.eval())









    



[[1 0 0 0]
 [0 0 2 0]
 [0 0 0 0]]
Check if this is identical with x:
 [[1 0 0 0]
 [0 0 2 0]
 [0 0 0 0]]



In [ ]: