tf.add(a, b) adds corresponding elements of a and b, applying NumPy-style broadcasting when the two tensors have compatible but different shapes — for example, adding a scalar to a tensor applies that scalar to every element, and adding a 1D tensor to a compatible 2D tensor applies it to every row. Using the + operator directly on tensors is equivalent and far more common in everyday TensorFlow code; tf.add() itself is mainly useful when you need to reference the operation explicitly, such as inside a computational graph.

1Understanding tf.add()

💡

Prefer the + operator over calling tf.add() directly for everyday code — they're functionally identical, and + is both more common and more readable.

editor.html

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([10, 20, 30])
print(tf.add(a, b))

localhost:3000

2Practical Example

Here is a real-world application of tf.add() showing how it is used in production TensorFlow code.

editor.html

import tensorflow as tf

matrix = tf.constant([[1, 2, 3], [4, 5, 6]])
row = tf.constant([10, 20, 30])
print(matrix + row)

localhost:3000

3Best Practices

Follow these guidelines when working with tf.add():

1. Use the + operator for everyday element-wise addition instead of calling tf.add() explicitly

2. Understand TensorFlow's broadcasting rules, the same as NumPy's, before relying on adding tensors of different shapes, so an unintended broadcast doesn't silently produce a wrong result

3. Check that both tensors' dtypes match before adding them, since TensorFlow, unlike NumPy, generally requires an explicit tf.cast() rather than silently upcasting mismatched types

⚠️

Tip: Prefer the + operator over calling tf.add() directly for everyday code — they're functionally identical, and + is both more common and more readable.

editor.html

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([10, 20, 30])
print(tf.add(a, b))

localhost:3000

Examples

Example 01Basic Usage

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([10, 20, 30])
print(tf.add(a, b))

Example 02Advanced Example

import tensorflow as tf

matrix = tf.constant([[1, 2, 3], [4, 5, 6]])
row = tf.constant([10, 20, 30])
print(matrix + row)

Best Practices

Use the + operator for everyday element-wise addition instead of calling tf.add() explicitly
Understand TensorFlow's broadcasting rules, the same as NumPy's, before relying on adding tensors of different shapes, so an unintended broadcast doesn't silently produce a wrong result
Check that both tensors' dtypes match before adding them, since TensorFlow, unlike NumPy, generally requires an explicit tf.cast() rather than silently upcasting mismatched types

Interview Question

Why does adding a tf.int32 tensor to a tf.float32 tensor raise an error in TensorFlow, when NumPy would silently upcast the result to float?▼

Hint: Think about how strict TensorFlow is about dtype mixing, compared to NumPy's automatic type promotion.

TensorFlow deliberately requires operations between tensors to have matching dtypes, raising an explicit error for a mismatch rather than silently upcasting one of them the way NumPy does — this is a design choice to catch potential bugs early and keep dtype handling explicit and predictable, especially important in a framework where dtype choices directly affect memory usage and hardware acceleration, like GPU/TPU support for specific precisions. To add tensors of different types together in TensorFlow, you need to explicitly convert one of them first with tf.cast(), making the type conversion a deliberate, visible step rather than an implicit, automatic one.

Exercises

MediumPractice using tf.add() in a real scenario.

View Solution

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([10, 20, 30])
print(tf.add(a, b))

Frequently Asked Questions

Why does adding a tf.int32 tensor to a tf.float32 tensor raise an error in TensorFlow, when NumPy would silently upcast the result to float?▼

Related Functions

tf-multiplytf-castnp-add