Each line of DataFrame is normal surgery. These functions are shyter: each line can be processed independently. With a multi-core CPE, many lines can be processed at the same time.
Until recently, exploit this opportunity opportunity was not possible. An application for a multilingual functioning, having a CPU-Bound, was thrown by the Global Uncrare Lock (Gil).
Python now provides a solution: “Free-Threating Ficting Create” YthonTon 3.13, Gil is removed, and the combination of true CPU is possible.
The work benefits are amazing. Python Freef Free-Threaded Python, Staticframe 3.2 may apply the wise work app in the Dataframe at least twice as a single murder.
For example, in each line of six million digit data square data, we can count the total amount of such prices lambda s: s.loc[s % 2 == 0].sum(). When using Python 3.13t (the “T” is a distinction included), time (measured with ipython %timeit) Decrease over 60%, from 21,3 MS to 7.89 MS:
# Python 3.13.5 experimental free-threading build (main, Jun 11 2025, 15:36:57) [Clang 16.0.0 (clang-1600.0.26.6)] on darwin
>>> import numpy as np; import static_frame as sf
>>> f = sf.Frame(np.arange(1_000_000).reshape(1000, 1000))
>>> func = lambda s: s.loc[s % 2 == 0].sum()
>>> %timeit f.iter_series(axis=1).apply(func)
21.3 ms ± 77.1 μs per loop (mean ± std. dev. of 7 runs, 10 loops each)
>>> %timeit f.iter_series(axis=1).apply_pool(func, use_threads=True, max_workers=4)
7.89 ms ± 60.1 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)Application to work intelligent in Staticframe uses iter_series(axis=1) Interface followed by the apply() (Through a single-impacted application) or apply_pool() with a lot of string (use_threads=True) or more processing (use_threads=False) Application.
Benefits of FREE PHYTHON FREE: Outperformance does not change throughout the DataFrame Shapes list and songs, equal to both macos and Linux, and a good scales for data size.
When using a typical Python with gil power-given gil, multiple-tied cfu strip processes often damage to work. As shown below, time to work the same in Standard Python Sanda from 17.7 MS by one thread at 40 MS for various characters:
# Python 3.13.5 (main, Jun 11 2025, 15:36:57) [Clang 16.0.0 (clang-1600.0.26.6)]
>>> import numpy as np; import static_frame as sf
>>> f = sf.Frame(np.arange(1_000_000).reshape(1000, 1000))
>>> func = lambda s: s.loc[s % 2 == 0].sum()
>>> %timeit f.iter_series(axis=1).apply(func)
17.7 ms ± 144 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
>>> %timeit f.iter_series(axis=1).apply_pool(func, use_threads=True, max_workers=4)
39.9 ms ± 354 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)There are off-offs when using the Python installed for free Python, usually, incorporates operations. This is the active Chython Development and EXFRESSION destination awaits 3.14T and beyond.
In addition, while many EC-extension packages are now offering binary wheelchairs that are previously included in 3.13t, the risks such as string.
Staticframe avoids these risks by enforcing unresponsive: rope safety, eliminated the requirement of the key or copies of defense. Staticframe does this through Numper Arty After (with flags.writeable Set to False) and prohibit local transformation.
Expanded DataFrame Testing
To view the useful features of the complex structure of data such as data data requires a number of types of datframes. The following performance panels enable the wise work app by 9rache gept with nine types of data, test all three cases and three levels of homogeneity.
With a limited number of components (eg a million), three shapes: long (10,000 100), square (100 at 10.0000). Variations of Homoogeneity, three phases of the data data are described: Columnar (no columns are closer to the same type), mixed (four closest columns, and all columns (all the same relative). The Staticframe allows for columns closely with the same type to represent two medicines, reducing the cost of column and line. In the ordinary, the whole Dataphrame can be displayed in one list of two. Data generation is generated with a framework for repair package.
The same activity is used: lambda s: s.loc[s % 2 == 0].sum(). While proper implementation is possible using direct Incess, this work is closest to regular applications where it is in the middle of many Series are created.
Legends legends for Councorrency Configuration documentation. When use_threads=TrueMany threads are used; when use_threads=FalseMulti-processing is used. StaticFrame uses ThreadPoolExecutor including ProcessPoolExecutor Meeting from the general library and disclose its parameters: max_workers Parameter describes the number of the upper strands or processes used. A chunksize The parameter is also available, but it is not the same in this study.
App with multiple three threaded machines with python 3.13t free
As shown below, multiple stringed processes in 3.13t are not consistent in all types of data test: Reduced speeds of at least 50%, and at other times more than 80%. The correct number of threads (the max_workers Parameter) is small for Tat Dataframes, as speedy processing of small lines means that extra artist looks over overwork.
Rate in the 100 objects of 100 objects (1E8), Outperformance improves. The processing period is reduced more than 70% of all other than two types of dataFam.
Multi-Threating higher than variations between platforms. In all cases, the brightness of using the free-installed Python is compatible with between the Macos and Linux, although Macos indicate serious benefits of years. Processing 100 million in Linux shows similar differences related to:
Amazingly, even the small dashaz of only ten thousand (1E4) components that can benefit from many threads in 3.13t. Although no profit is available for broad dataframes, the long-term Fatafadrames process can be reduced in half.
The app with many compiles combined with regular Python 3.13
Before the Python is free of charge, many CPU struggling processing has led to unprotected performance. This is done below, where similar tests are being conducted with the standard Python 3.13.
Multi-Secured Funct request with Python 3.13 Normal
Before the Python is free of charge, much processing was the only CPU-bound of Confull. However, processing many benefits only if the amount of each procedure is sufficient for preparing high costs to create each translator process and copy data between procedures.
As indicated here, the most intelligent work app is the most beautiful to disproves the effectiveness, time processing increases from two to 10 times long-length of the length of the tle -Jut. Each unit of work is too small to work over the surface processing.
Python's Paython status is free
PEP 703, “to make a global translator in Chython”, is accepted by Python Steering Council in 2023 in the signature of the first phase (Python 3.13) to evaluate and impersonate; In the second phase, it becomes an exploring and legally supported; In the third category, it becomes the use of the default Python.
After the development of important chython, and the support of sensitive packages such as NUMPEL, PEP 779, “Support Methods Stekering Council in 2025.
Store
Application for wise performance is just the beginning: A group-by works, the Window Funct app, and many other approachable data functions are well matched with the same performance question.
The work of making chsthon quickly has a success: Python 3.14 is said to be 20% to 40% as soon as Python 3.10. Unfortunately, those applicable benefits were not available for multiple functional Dafeframes, where the work was arrested within the C-Extensions (can be inpupw, aforfylist).
As shown, FREE Python is enabled the corresponding rating slain using the most expensive fibers, working with memory, bringing 50% to the reduction of libraries, whether to perform CD libraries. Empowerment of safety with safe data structures in all fibers, opportunities to improve greater performance now.
