TensorFlow | TF tensor

In the previous article, we learnt how to use Google Colaboratory to get started with TensorFlow. In this tutorial we will cover tensors in TensorFlow.

Introduction to Tensors

Tensors are multi-dimensional arrays having a uniform type that is called a dtype.

Similar to Python strings and numbers, tensors are immutable. A tensor's contents can never be changed, only a new tensor can be created.

How to create a tensor

Let's get started with creating a tensor in TensorFlow. We would be using Google Colaboratory to write the code, if you don't know how to use Google Colaboratory refer tensorflow with google colab to get started.

Create a New notebook in Colaboratory. Write below code in the cell and run the cell.

   
 
 import tensorflow as tf
 t1 = tf.constant(6)
 t1    
    

Output

Properties of a tensor

• a data type - tensor t1 is having data type as int32
• a shape - tensor t1 is having shape t1.shape, that means its a scalar or rank 0 tensor.
• numpy function (available in eager execution) - numpy function associated with tensor returns A NumPy array of the same shape and dtype.

Tensor DType's

dtype property of Tensor represents the type of the elements in a Tensor. Following are some of the DType's defined in TensorFlow.

For a complete list of DType's refer this link.

• bfloat16
• float16
• float32
• float64
• int16
• int32
• int64
• int8
• string
• uint16
• uint32
• uint64
• uint8

Tensor DType's attributes

Let's have a look to some os the important methods available in tf.dtypes module.

tf.dtype.is_compatible_with()

tf.dtype.is_compatible_with is used to check if a Tensor of the `other` `DType` will be implicitly converted to this `DType`. To understand this have a look to the below code snippet.

   
   import tensorflow as tf
   import numpy as np
   
   t1 = tf.constant(6, dtype=np.int16)
   t2 = tf.constant(6, dtype=tf.int16)
   t3 = tf.constant(6.0, dtype=tf.float16)
   
   print(t1)
   print(t2)
   print(t3)
 
   print(t1.dtype.is_compatible_with(t2.dtype))
   
   ptint(t2.dtype.is_compatible_with(t3.dtype))
  

Output

tf.dtype.limits

tf.dtype.limits returns the tuple containing min, max intensity limits of dtype.

   
   import tensorflow as tf
 
   t1 = tf.constant(6)
   print(t1)
   
   print(t1.dtype.limits)
   
  

Output

tf.dtype.is_integer

tf.dtype.is_integer returns True if the dtype of tensor is (non-quantized) integer type.

   
   import tensorflow as tf
 
   t1 = tf.constant(6)
   print(t1.dtype.is_integer)
  

tf.dtype.is_floating

tf.dtype.is_floating returns True if the dtype of tensor is (non-quantized, real) floating point type.

   
   import tensorflow as tf
 
   t1 = tf.constant(6.0)
   print(t1.dtype.is_floating)
  

tf.dtype.is_bool

tf.dtype.is_bool returns True if the dtype of tesnor is a boolean data type.

   
   import tensorflow as tf
 
   t1 = tf.constant(True, dtype=tf.bool)
   print(t1.dtype.is_bool)
   
   t2 = tf.constant(False, dtype=tf.bool)
   print(t2.dtype.is_bool)