PyTorch .div()

The .div() function performs element-wise division between tensors or divides a tensor by a scalar value. It is a fundamental tensor operation in PyTorch, commonly used for performing machine learning and deep learning tasks such as data preprocessing, normalization, and optimization.

Element-wise operations are essential in tensor computations, enabling efficient parallel processing. The .div() function offers a simple and optimized way to handle division across tensors in neural networks and mathematical transformations.

Syntax

torch.div(input, other, *, rounding_mode=None, out=None)

Parameters:

• input: The input tensor (dividend).
• other: The tensor or scalar to divide by (divisor).
• rounding_mode (Optional): Controls the rounding behavior. Can be None (default), trunc, or floor.
• out (Optional): The output tensor to store the result.

Return value:

A tensor with the result of element-wise division. If out is provided, the returned tensor is the same as out.

Example 1: Basic Usage of .div() with Tensors

This example demonstrates how to use .div() to perform element-wise division between two tensors of the same shape:

import torch

# Create input tensors
tensor1 = torch.tensor([4.0, 9.0, 16.0, 25.0])
tensor2 = torch.tensor([2.0, 3.0, 4.0, 5.0])

# Perform element-wise division
result = torch.div(tensor1, tensor2)

# Print results
print("Tensor1 (dividend):", tensor1)
print("Tensor2 (divisor):", tensor2)
print("Division result:", result)

Copy to clipboard
Copy to clipboard

Here is the output:

Tensor1 (dividend): tensor([ 4.,  9., 16., 25.])
Tensor2 (divisor): tensor([2., 3., 4., 5.])
Division result: tensor([2., 3., 4., 5.])

Copy to clipboard
Copy to clipboard

The .div() operation computes the element-wise division of tensor1 by tensor2. Each element in the resulting tensor is the quotient of the corresponding elements in the input tensors.

Example 2: Division with Rounding Modes

The .div() function supports optional rounding modes when performing integer division, which control how the result is rounded:

import torch

# Create input tensors
tensor1 = torch.tensor([5, 7, 10, 15])
tensor2 = torch.tensor([2, 2, 3, 4])

# Division with different rounding modes
result_default = torch.div(tensor1, tensor2)
result_floor = torch.div(tensor1, tensor2, rounding_mode='floor')
result_trunc = torch.div(tensor1, tensor2, rounding_mode='trunc')

# Print results
print("Default division:", result_default)
print("Floor division:", result_floor)
print("Trunc division:", result_trunc)

Copy to clipboard
Copy to clipboard

Here is the output:

Default division: tensor([2.5000, 3.5000, 3.3333, 3.7500])
Floor division: tensor([2, 3, 3, 3])
Trunc division: tensor([2, 3, 3, 3])

Copy to clipboard
Copy to clipboard

The rounding modes control how the division results are handled:

• Default mode performs true division, returning floating-point results.
• floor mode rounds the result down toward negative infinity (i.e., floor division).
• trunc mode truncates the decimal part, rounding toward zero.

Example 3: Data Normalization with .div()

The .div() function is commonly used in data preprocessing and normalization. Here’s an example of normalizing a dataset:

import torch

# Create a sample dataset (3 samples with 4 features each)
dataset = torch.tensor([
  [10.0, 20.0, 30.0, 40.0],
  [15.0, 25.0, 35.0, 45.0],
  [20.0, 30.0, 40.0, 50.0]
])

# Calculate the range (max - min) for each feature
feature_min = dataset.min(dim=0).values
feature_max = dataset.max(dim=0).values
feature_range = feature_max - feature_min

# Normalize the dataset to [0, 1] range using div
normalized_data = torch.div(dataset - feature_min, feature_range)

print("Original dataset:")
print(dataset)
print("\nNormalized dataset:")
print(normalized_data)

Copy to clipboard
Copy to clipboard

Here is the output:

Original dataset:
tensor([[10., 20., 30., 40.],
        [15., 25., 35., 45.],
        [20., 30., 40., 50.]])

Normalized dataset:
tensor([[0.0000, 0.0000, 0.0000, 0.0000],
        [0.5000, 0.5000, 0.5000, 0.5000],
        [1.0000, 1.0000, 1.0000, 1.0000]])

Copy to clipboard
Copy to clipboard

This example demonstrates how .div() can be used to normalize data to a specific range, which is a commonly performed preprocessing step in machine learning workflows.