1. Introduction
PyPy is a faster version of python
- The package used is exactly the same and the use method is exactly the same.
- However, not all packages in python are supported.
- Yes, for Python 3 7, Python2. 7.
- Install pypy3 After 7, use the PY3 command
- Install pypy2 After 7, use the py2 command
- performance
- according to Introduction to pypy: introduction to the installation and use of pypy , PyPy is 17 times faster than Python
- However, it is not necessarily faster. I use PY3 7 and python 3 8 run, I wrote the optical tweezers program , it is found that the former takes 0.436s and the latter takes 0.283s
2. PyPy installation configuration under Windows
- Step 1: create a virtual environment conda create -n pypy_env
- Step 2: activate the virtual environment_ env. You can see that under the virtual environment directory (C:\Users \ username \. conda\envs\pypy_env), there are only Scripts and conda meta folders, and the content is only conda
- Step 3: unzip and install. Pypy official website download , after downloading, unzip it and copy the content to the virtual environment directory. If you do not need a virtual environment, neither of the first two steps is required. Just add the PyPy folder to the Path environment variable
- Under windows, you should also ensure that VC runtime library is installed Pypy's official download page has a link to provide its installation
Then install the pip tool pypy3 -m ensurepip
After installing the package, you can directly use the command pypy3 -m pip install package name.
- For example, PY3 - M PIP install numpy
reference resources:
- PY3 installation configuration under windows
- Running python with Windows + PyPy+Sublime/PyCharm: improve the running speed of Python
3. Special package
- numpy,imageio: no problem
- Matplotlib: when the version is not specified, the installation always fails. When installing versions 2.2 and 3.3.1, it is successful. pypy3 -m pip install matplotlib==3.3.1
- pandas: the installation always fails when the version is not specified. When version 1.2.0 is installed, it is successful. pypy3 -m pip install matplotlib==1.2.0
- scipy: direct installation will fail. The installation is slow because it compiles the scipy package, which is very large
4. Test
According to the following two tests, the
- When executing loops, PyPy is dozens of times faster than Python
- In the use of other libraries, the speed is not necessarily faster. Even slower. Such as numpy
- numpy is written in C + +. So its execution has been very fast. If you use pypy, it won't get faster. Similarly, SciPy and pandas should also be the test results
4.1. Test cycle
import time
t = time.time()
for i in range(10**8):
continue
print(time.time() - t)
Test results, time consuming:
- python3.8: 2.956s
- PY3: 0.060s (33 times faster)
4.2. numpy test
import numpy as np import time t = time.time() # row is y, column is x row_count = 4024 col_count = row_count center_x = col_count//2 - 0.5 center_y = row_count//2 - 0.5 x0 = np.arange(col_count) y0 = np.arange(row_count) x0 = x0 - center_x # Positive to the right y0 = center_y - y0 # Positive upward x_grid, y_grid = np.meshgrid(x0, y0) half_size = row_count / 2 x = x_grid.flatten() / half_size # Converted to values between [- 1,1], unit circle y = y_grid.flatten() / half_size x2 = x ** 2 y2 = y ** 2 x3 = x ** 3 y3 = y ** 3 x4 = x ** 4 y4 = y ** 4 xy = x * y Z = np.zeros([15, len(x)]) # Z0 # Fun1 = lambda x, y: 1 Z[0, :] = np.ones_like(x) # Z1 # Fun1 = lambda x, y: y Z[1, :] = y # Z2 # Fun2 = lambda x, y: x Z[2, :] = x # Z3 # Fun3 = lambda x, y: 2*x*y Z[3, :] = 2 * xy # Z4 # Fun4 = lambda x, y: -1 + 2 * (x ** 2 + y ** 2) Z[4, :] = -1 + 2 * (x2 + y2) # Z5 # Fun5 = lambda x, y: x**2 - y**2 Z[5, :] = x2 - y2 # Z6 # Fun6 = lambda x, y: 3 * x**2 * y - y**3 Z[6, :] = 3 * x2 * y - y3 # Z7 # Fun7 = lambda x, y: -2*y + 3*y * (x**2 + y**2) # It doesn't match the effect drawing Z[7, :] = -2 * y + 3 * y * (x2 + y2) # Z8 # Fun8 = lambda x, y: -2*x + 3*x * (x**2 + y**2) # It doesn't match the effect drawing Z[8, :] = -2 * x + 3 * x * (x2 + y2) # Z9 # Fun9 = lambda x, y: x**3 - 3*x*y**2 Z[9, :] = x3 - 3 * x * y2 # Z10 # Fun10 = lambda x, y: 4 * x**3 * y - 4 * x * y**3 Z[10, :] = 4 * x3 * y - 4 * x * y3 # Z11 # Fun11 = lambda x, y: -6*x*y + 8*x*y * (x**2 + y**2) # It doesn't match the effect drawing Z[11, :] = -6 * xy + 8 * xy * (x2 + y2) # Z12 # Fun12 = lambda x, y: 1 - 6*(x**2 + y**2) + 6*(x**2 + y**2)**2 # It doesn't match the effect drawing Z[12, :] = 1 - 6 * (x2 + y2) + 6 * (x2 + y2)**2 # Z13 # Fun13 = lambda x, y: -3*x**2 + 3*y**2 + 4*x**4 - 4*y**4 # It doesn't match the renderings I deduced that its formula is correct, but the coefficient root sign 10 is missing Z[13, :] = -3 * x2 + 3*y2 + 4 * x4 - 4 * y4 # Z14 # Fun14 = lambda x, y: x**4 - 6 * x**2 * y**2 + y**4 Z[14, :] = x4 - 6 * x2 * y2 + y4 print(time.time() - t)
Test results, time consuming:
- python3.8: 6.092s
- PY3: 6.195s (slightly slower)