TensorFlow is an open source software library for high-performance numerical computation.

Tensors are arrays of arbitrary dimension. TensorFlow can be used to manipulate Tensors of very high dimension.

Some low dimensional Tensors are:

A scalar is a 0-d array (a 0th-order tensor). For example, "Howdy" or 5
A vector is a 1-d array (a 1st-order tensor). For example, [2, 3, 5, 7, 11] or [5]
A matrix is a 2-d array (a 2nd-order tensor). For example, [[3.1, 8.2, 5.9][4.3, -2.7, 6.5]]

A TensorFlow graph(computational graph, data flow graph) is a Graph Data Structure. The graph's nodes are operations. Edges are Tensors. Tensors flow through a graph, manipulated at each node by an operation.

Tensors are stored in graphs as constants or Variables.

import tensorflow as tf

# Tensors can be stored in graph as constants or Variables.
x = tf.constant(5.2)
y = tf.Variable([5])
print(x,y)

# Assign a diffetent value
y = y.assign([6])

# Graphs run within a TensorFlow Session, which holds the state for the graph(s)

with tf.Session() as sess:
initialization = tf.global_variables_initializer()
print(y.eval())

# When working eith tf.Variables, you must explicitly initialize them by calling tf.global_variables_initializer() at the start of the session

# After defining these you can cobine them with other operations like tf.add. A new tensor will be returned with the sum of these 2

# create a graph
g = tf.Graph()

# Establish the graph as default graph
with g.as_default():
   x = tf.constant(8, name='x_const')
   y = tf.constant(5, name='y_const')
   z = tf.constant(4, name='z_const')
   my_sum = tf.add(x, y, name='x_y_sum')
   my_sum_z = tf.add(my_sum, z, name='x_y_z_sum')
   # create a session to run the default graph
   with tf.Session() as sess:
       print(my_sum.eval())
       # Output: 13
       print(my_sum_z.eval())
       # Output: 17

# Working with Vectors
print('VECTOR ADDITION')
with tf.Graph().as_default():
   # create 6 element vectors
   primes = tf.constant([2,3,5,7,11,13], dtype=tf.int32)
   ones = tf.ones([6], dtype=tf.int32)   # [1,1,1,1,1,1]
   # Add them n store in another vector
   my_sum = tf.add(primes, ones)

   with tf.Session() as sess:
       print('Vector Sum: ', my_sum.eval())
       # ('Vector Sum: ', array([ 3, 4, 6, 8, 12, 14], dtype=int32))

print('TENSOR SHAPES')
# shapes categorize size and number of dimensions in a tensor. Its a list, with ith element representing size along dimension i. Length of this list indicates rank of tensor(no. of dimensions)

with tf.Graph().as_default():
   # 0-D Tensor
   scalar = tf.zeros([])
   # 3 elements
   vector = tf.zeros([3])
   # 2 rows, 3 columns
   matrix = tf.zeros([2,3])

   with tf.Session() as sess:
       print('scalar shape: ', scalar.get_shape(), 'value: ', scalar.eval())
       print('vector shape: ', vector.get_shape(), 'value: ', vector.eval())
       print('matrix shape: ', matrix.get_shape(), 'value: ', matrix.eval())

print('BROADCASTING')
# A smaller array is enlarged to ahve the same shape as larger array to perform element wise operation

with tf.Graph().as_default():
   primes = tf.constant([2,3,5,7,11,13], dtype=tf.int32)
   ones = tf.constant(1, dtype=tf.int32)
   my_sum = tf.add(primes, ones)
   with tf.Session() as sess:
       print(my_sum.eval())       # [ 3 4 6 8 12 14]

print('MATRIX MULTIPLICATION')
# In linear algebra, nc of first matrix shud be equal to nr in second for Matrix multiplication.
with tf.Graph().as_default():
   # 3 * 4 2-d tensor
   x = tf.constant([[1,2,3, 4], [5,6,7,8], [9,10,11,12]], dtype=tf.int32)
   # 4*2
   y = tf.constant([[2,2], [3,5], [4,5], [1,5]], dtype=tf.int32)
   my_mul = tf.matmul(x,y)
   with tf.Session() as sess:
       print(my_mul.eval())
   # [[ 24 47]
    # [ 64 115]
    # [104 183]]

print('TENSOR RESHAPING')
with tf.Graph().as_default():
   # 8 * 2 matrix (2-D tensor)
   matrix = tf.constant([[1,2], [3,4], [5,6], [7,8],
                       [9,10], [11,12], [13,14], [15,16]], dtype=tf.int32)
   reshaped_2_8 = tf.reshape(matrix, [2,8])
   # reshape to 3-d
    reshaped_2_2_8 = tf.reshape(matrix, [2,2,4])
   with tf.Session() as sess:
       print('reshape to 2 * 8')
       print(reshaped_2_8.eval())
       # [[ 1 2 3 4 5 6 7 8]
        # [ 9 10 11 12 13 14 15 16]]
        print('reshape to 2 * 2 * 4')
        print(reshaped_2_2_8.eval())
        # [[[ 1 2 3 4]
# [ 5 6 7 8]]

# [[ 9 10 11 12]
# [13 14 15 16]]]

print('Variables')
# Values can be changed
with tf.Graph().as_default():
   v = tf.Variable([3])
   w = tf.Variable(tf.random_normal([1], mean=1.0, stddev=0.35))
   with tf.Session() as sees:
       try:
           print(v.eval())
       except tf.errors.FailedPreconditionError as e:
           print(e)
           # Error that variable is uninitialized
       sees.run(tf.global_variables_initializer())
       print(v.eval())   # [3]
       print(w.eval())   # [0.98439926]
# with every new session, variable values have to be reinitialized
       assignment = tf.assign(v, [7])
       print('reassigned. wont chnge: ',v.eval())   # Value will not change until reinitialized
       sees.run(assignment)
       print('reinitialized. value changes: ',v.eval())

Siguiente

Anterior

Notes on Prerequisites