In this chapter, you will get introduced to tensors and their operations in TensorFlow 2.0. A tensor is a vector (or matrix) of n-dimensions where all elements have the same data type. In other words, a tensor is a collection of feature vectors (i.e., arrays) of n-dimensions. Some of the examples of tensors are:
$\left[[ 1] \ \ [ 2]\right]$
$\left[[ 1 \ \ 2] \ [ 3 \ \ 4]\right]$
$\left[[ 1 \ \ 2 \ \ 3] \ [ 4 \ \ 5 \ \ 6]\right]$
TensorFlow 2.0 supports the following 4 types of tensors: Variable, Constant, Placeholder and SparseTensor. Tensor objects will have the following three properties:
Except for the Variable, all other tensor types are immutable, i.e., their value will not change during opertions. Here, we will focus primarily on the Variable and Constant tensors.
While creating a tensor object, we need to specify the Data and Data Type.
var = tf.Variable([1, 2], tf.int32) # Variable Tensor of data type int32 const = tf.constant([1.1, 2.2], tf.float64) # Constant Tensor of data type float64 print(f'Variable: {var}') # Print Variable print(f'Contant: {const}') # Print Constant print(f'\nData of Variable: {var.numpy()}') # Print Data of Variable as NumPy array print(f'Data of Constant: {const.numpy()}') # Print Data of Constant as NumPy array print(f'\nShape of Variable: {var.shape}') # Print Shape of Variable as a tuple print(f'Shape of Constant: {const.shape}') # Print Shape of Constant as a tuple print(f'\nData type of Variable: {var.dtype}') # Print data type of Variable print(f'Data type of Constant: {const.dtype}') # Print data type of Constant
Variable: <tf.Variable 'Variable:0' shape=(2,) dtype=int32, numpy=array([1, 2], dtype=int32)> Contant: [1.1 2.2] Data of Variable: [1 2] Data of Constant: [1.1 2.2] Shape of Variable: (2,) Shape of Constant: (2,) Data type of Variable: <dtype: 'int32'> Data type of Constant: <dtype: 'float64'>
Various arithmetic operations such as addition, subtraction, multiplication, etc. can be performed on tensors.
# Taking two tensors of type int32 tensor_a = tf.Variable([[1,2]], dtype = tf.int32) tensor_b = tf.Variable([[3, 4]], dtype = tf.int32) # Arithmetic Operations tensor_add = tf.add(tensor_a, tensor_b) # a + b print(f'Addition: {tensor_add}') tensor_sub = tf.subtract(tensor_b, tensor_a) # b - a print(f'Subtraction: {tensor_sub}') tensor_mul = tf.multiply(tensor_a, tensor_b) # a * b print(f'Multiplication: {tensor_mul}') tensor_div = tf.divide(tensor_a, tensor_b) # a / b print(f'Division: {tensor_div}') tensor_pow = tf.pow(tensor_a, tensor_b) # a ^ b print(f'Power: {tensor_pow}')
Addition: [[4 6]] Subtraction: [[2 2]] Multiplication: [[3 8]] Division: [[0.33333333 0.5 ]] Power: [[ 1 16]]
Tensors can also be reshaped similar to NumPy arrays.
tensor_a = tf.ones((6 ,6), dtype = tf.int32) # Create a tensor of shape (6, 6) with all values 1 print(f'Initial Tensor:\n {tensor_a.numpy()}') print(f'Initial Shape: {tensor_a.shape}') # Reshape the tensor into shape (9, 4) reshaped_a = tf.reshape(tensor_a, (9, 4)) print(f'\nReshaped Tensor:\n {reshaped_a.numpy()}') print(f'Reshaped Shape: {reshaped_a.shape}')
Initial Tensor: [[1 1 1 1 1 1] [1 1 1 1 1 1] [1 1 1 1 1 1] [1 1 1 1 1 1] [1 1 1 1 1 1] [1 1 1 1 1 1]] Initial Shape: (6, 6) Reshaped Tensor: [[1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1] [1 1 1 1]] Reshaped Shape: (9, 4)
This is it for the basics of Tensors and Tensor Operations in TensorFlow 2.0. In the next chapter, you will learn to perform Linear Regressing using TensorFlow 2.0.
Do you want to learn Python, Data Science, and Machine Learning while getting certified? Here are some best selling Datacamp courses that we recommend you enroll in: