tf.multiply() multiplies two tensors element-wise, and is exactly what the * operator calls on TensorFlow tensors — distinct from matrix multiplication.

Syntax

tf.multiply(x, y)
x * y

Deep Dive Course

tf.multiply(a, b) multiplies corresponding elements of a and b, applying broadcasting as needed — this is element-wise multiplication, sometimes called the Hadamard product, not matrix multiplication, which uses the separate tf.matmul() function or the @ operator instead. Mixing these two up, expecting * to perform matrix multiplication, is one of the most common sources of subtle shape-related bugs when translating mathematical notation into TensorFlow code.

1Understanding tf.multiply()

💡

Never use * expecting matrix multiplication in TensorFlow — it always multiplies element-wise, broadcasting shapes as needed; use tf.matmul() or the @ operator specifically when you mean actual matrix multiplication.

editor.html

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([4, 5, 6])
print(tf.multiply(a, b))

localhost:3000

2Practical Example

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

editor.html

import tensorflow as tf

values = tf.constant([10, 20, 30, 40])
mask = tf.constant([1, 0, 1, 0])
print(values * mask)

localhost:3000

3Best Practices

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

1. Use * (or tf.multiply()) only for element-wise multiplication, and tf.matmul() or @ for true matrix multiplication — never assume they're interchangeable

2. Double-check tensor shapes before relying on broadcasting in a multiplication, since a shape mismatch can silently produce an unexpected result rather than always raising an error

3. Use element-wise multiplication with a 0/1 mask tensor as a common technique to selectively zero out or keep specific elements of another tensor

⚠️

Tip: Never use * expecting matrix multiplication in TensorFlow — it always multiplies element-wise, broadcasting shapes as needed; use tf.matmul() or the @ operator specifically when you mean actual matrix multiplication.

editor.html

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([4, 5, 6])
print(tf.multiply(a, b))

localhost:3000

Examples

Example 01Basic Usage

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([4, 5, 6])
print(tf.multiply(a, b))

Example 02Advanced Example

import tensorflow as tf

values = tf.constant([10, 20, 30, 40])
mask = tf.constant([1, 0, 1, 0])
print(values * mask)

Best Practices

Use * (or tf.multiply()) only for element-wise multiplication, and tf.matmul() or @ for true matrix multiplication — never assume they're interchangeable
Double-check tensor shapes before relying on broadcasting in a multiplication, since a shape mismatch can silently produce an unexpected result rather than always raising an error
Use element-wise multiplication with a 0/1 mask tensor as a common technique to selectively zero out or keep specific elements of another tensor

Interview Question

What's the difference between A * B and tf.matmul(A, B) for two 2D tensors?▼

Hint: Think about element-wise multiplication versus true matrix multiplication.

A * B performs element-wise multiplication, requiring A and B to have the same shape, or shapes compatible via broadcasting, multiplying corresponding elements together independently. tf.matmul(A, B) performs true matrix multiplication, following the standard linear algebra rule where the number of columns in A must match the number of rows in B, computing dot products of rows and columns rather than simply multiplying elements in place. Confusing the two, especially when coming from mathematical notation where juxtaposition usually implies matrix multiplication, is a very common source of bugs in TensorFlow code.

Exercises

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

View Solution

import tensorflow as tf

a = tf.constant([1, 2, 3])
b = tf.constant([4, 5, 6])
print(tf.multiply(a, b))

Frequently Asked Questions

What's the difference between A * B and tf.matmul(A, B) for two 2D tensors?▼

Related Functions

tf-addtf-matmulnp-multiply