Comparison of operation functions of PyTorch tensor and NumPy array

PyTorch and NumPy are two widely used libraries in deep learning and scientific computing.

Although they both provide the functionality of array operations, there are some differences between them.

This article will compare the similarities and differences between PyTorch tensors and NumPy arrays in terms of arithmetic functions to help readers better understand how to use them.

Differences in basic array operations.

First, let's look at their differences in basic array operations.

The NumPy array is a collection of multidimensional arrays, while the PyTorch tensor is a continuous multidimensional array.

Therefore, when using certain arithmetic functions, the behavior of the two may be different.

For example, during addition operations, NumPy arrays are automatically broadcast (broadcasting), while PyTorch tensors need to manually specify broadcast rules.

\n#

Example: addition.

import numpy as np
import torch

# NumPy 自动广播
a = np.array([1, 2, 3])
b = np.array([[1], [2], [3]])
c = a + b  # 结果为 [[2, 3, 4], [3, 4, 5], [4, 5, 6]]

# PyTorch 需要手动指定广播规则
a_torch = torch.tensor([1, 2, 3])
b_torch = torch.tensor([[1], [2], [3]])
c_torch = a_torch + b_torch  # 结果为 [[2, 3, 4], [3, 4, 5], [4, 5, 6]]

Differences in advanced array operations.

Second, let's look at their differences in advanced array operations.

NumPy provides some advanced functions such as numpy.fftFor Fourier transform, PyTorch provides its own FFT module ( torch.fft)。

In addition, NumPy also supports parallel computing, enabling multi-threaded acceleration by setting environment variables, while PyTorch uses CUDA for GPU acceleration.

\n#

Example: Fourier transform.

import numpy as np
import torch

# NumPy FFT
x = np.array([1, 2, 3, 4])
y = np.fft.fft(x)

# PyTorch FFT
x_torch = torch.tensor([1, 2, 3, 4])
y_torch = torch.fft.fft(x_torch)

Differences in data types.

Finally, we also need to note the difference in data types between the two.

NumPy supports a variety of data types, including integers, floating-point numbers, boolean values, etc., while PyTorch mainly supports floating-point numbers (float32 and float64).

In addition, since PyTorch is a dynamic graph-based framework, it also provides some special data types, such as torch.tensorUsed to create tensor objects.

\n#

Example: data type conversion.

import numpy as np
import torch

# NumPy 支持多种数据类型
a = np.array([1, 2, 3], dtype=np.int32)
b = np.array([1.0, 2.0, 3.0], dtype=np.float32)
c = np.array([True, False, True], dtype=np.bool)

# PyTorch 主要支持浮点数类型
a_torch = torch.tensor([1, 2, 3], dtype=torch.int32)
b_torch = torch.tensor([1.0, 2.0, 3.0], dtype=torch.float32)
c_torch = torch.tensor([True, False, True], dtype=torch.bool)

Summarize.

Although both PyTorch and NumPy provide the ability to operate arrays, they differ in some ways.

For developers, understanding these differences can help them more effectively choose the data structures and algorithms that suit their projects.

In practical application scenarios, choosing the right tool can greatly improve development efficiency and code performance.