cast() creates a new tensor with the same values, converted as needed, and shape as the input, but with a different dtype — converting a float to an int truncates any fractional part, the same truncation behavior as Python's int(), rather than rounding, and converting between different floating-point precisions, like float64 to float16, can lose precision if the new type can't represent the original value's full accuracy. This is commonly needed because TensorFlow, like NumPy, requires every element in a tensor to share exactly one dtype, and different layers or operations sometimes expect specific, particular types.

1Understanding tf.cast()

💡

tf.cast() truncates when converting a float to an integer type, the same as Python's int(), rather than rounding — use tf.round() first if you specifically want conventional rounding behavior before converting to an integer type.

editor.html

import tensorflow as tf

x = tf.constant([1.7, 2.3, 3.9])
casted = tf.cast(x, tf.int32)
print(casted)

localhost:3000

2Practical Example

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

editor.html

import tensorflow as tf

x = tf.constant([1, 0, 1, 1], dtype=tf.int32)
as_bool = tf.cast(x, tf.bool)
print(as_bool)

localhost:3000

3Best Practices

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

1. Cast input data to the dtype a specific layer or operation expects, rather than assuming automatic type coercion will happen

2. Use tf.round() before casting to an integer type if you need rounding rather than the truncation cast() performs by default

3. Be aware that casting to a lower-precision type, like float16, can lose accuracy — verify this is an acceptable tradeoff, typically for memory/speed optimization, before doing so

⚠️

Tip: tf.cast() truncates when converting a float to an integer type, the same as Python's int(), rather than rounding — use tf.round() first if you specifically want conventional rounding behavior before converting to an integer type.

editor.html

import tensorflow as tf

x = tf.constant([1.7, 2.3, 3.9])
casted = tf.cast(x, tf.int32)
print(casted)

localhost:3000

Examples

Example 01Basic Usage

import tensorflow as tf

x = tf.constant([1.7, 2.3, 3.9])
casted = tf.cast(x, tf.int32)
print(casted)

Example 02Advanced Example

import tensorflow as tf

x = tf.constant([1, 0, 1, 1], dtype=tf.int32)
as_bool = tf.cast(x, tf.bool)
print(as_bool)

Best Practices

Cast input data to the dtype a specific layer or operation expects, rather than assuming automatic type coercion will happen
Use tf.round() before casting to an integer type if you need rounding rather than the truncation cast() performs by default
Be aware that casting to a lower-precision type, like float16, can lose accuracy — verify this is an acceptable tradeoff, typically for memory/speed optimization, before doing so

Interview Question

Why does casting the float value 2.3 to an integer type with tf.cast() produce 2, the same truncated result you'd also get for a value like 2.9?▼

Hint: Think about truncation versus rounding as two different, distinct operations.

tf.cast() specifically truncates the fractional part of a float when converting to an integer type, simply discarding everything after the decimal point regardless of whether it's closer to the number above or below, the same truncation behavior as Python's own int() function. This is a fundamentally different operation from rounding, which considers how close the value is to each neighboring integer and picks the nearest one — cast() never performs that comparison at all, so a value close to the next integer still truncates down rather than rounding up, which surprises people expecting rounding behavior instead.

Exercises

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

View Solution

import tensorflow as tf

x = tf.constant([1.7, 2.3, 3.9])
casted = tf.cast(x, tf.int32)
print(casted)

Frequently Asked Questions

Why does casting the float value 2.3 to an integer type with tf.cast() produce 2, the same truncated result you'd also get for a value like 2.9?▼

Related Functions

tf-reshapenp-arrayint()