Bilingual original link: Top Python Libraries for Deep Learning, Natural Language Processing & Computer Vision
Please note that the following icon is drawn by Gregory Piatetsky , each library has its category, drawn by star and contributors, and the symbol size is represented by the logarithm of the number of commits of the library on Github.
Figure 1: Top Python libraries for deep learning, natural language processing and computer vision
Draw by star rating and number of contributors; logarithm of the number of submissions indicates relative size
Without further ado, here are 30 top Python libraries carefully selected by KDnuggets staff that can be used for deep learning, natural language processing and computer vision.
Deep learning
1. TensorFlow
Star: 149000, submissions: 97741, contributors: 754
TensorFlow is an end-to-end open source platform for machine learning. It has comprehensive and flexible tools, libraries and community resources that can help researchers promote the development of advanced machine learning technologies and developers more easily develop and publish applications supported by machine learning.
2. Hard
Star: 50000, number of submissions: 5349, contributors: 864
Keras is a machine learning API written by python, which runs on TensorFlow, the top platform for machine learning.
3. PyTorch
Star: 43200, number of submissions: 30696, contributors: 1619
Tensors and dynamic neural networks implemented in Python with powerful GPU acceleration.
4. employment
Star: 19800, submissions: 1450, contributors: 607
By using the best technical practices at the moment, fastai simplifies the training process extremely quickly and accelerates the neural network.
5. PyTorch Lightning
Star: 9600, submissions: 3594, contributors: 317
A packaged lightweight version of PyTorch for high-performance AI research. You can shrink your model instead of providing a small model.
6. JAX
Star: 10000, number of submissions: 5708, contributors: 221
Combination transformation of Python+NumPy programs: differentiation, vectorization, JIT on GPU/TPU, etc.
7. MXNet
Star: 19100, number of submissions: 11387, contributors: 839
A lightweight, convenient and flexible distributed/mobile machine learning library with dynamic and mutation-aware data flow management scheduler: supports Python, R, Julia, Scala, Go, JavaScript, etc.
8. Ignite
Star: 3100, submissions: 747, contributors: 112
A high-level library for training and evaluating PyTorch neural networks flexibly and transparently.
Natural language processing
9. FastText
Star: 21700, number of submissions: 379, contributors: 47
fastText is a library for efficiently learning word ideograms and sentence classification.
10. spaCy
Star: 17400, number of submissions: 11628, contributors: 482
Industrial-grade natural language processing library (NLP) implemented using Python and Cython
11. gensim
Star: 11,200, submissions: 4024, contributors: 361
gensim uses a large corpus for topic modeling, document indexing and similarity retrieval. The target audience is the natural language processing (NLP) and information retrieval (IR) communities
12. NLTK
Star: 9300, submissions: 13990, contributors: 319
NLTK-Natural Language Toolbox-is a complete set of open source Python modules, data sets and tutorials for natural language processing research and development
Star: 4300, number of submissions: 568, contributors: 64
Fast, efficient, open data sets and evaluation indicators using PyTorch, TensorFlow, NumPy and Pandas tools for natural language processing
Star: 3800, number of submissions: 1252, contributors: 30
Fast top word segmenter for research and production
Star: 3500, number of submissions: 5480, contributors: 585
Transformers: The top natural language processing library for Pytorch and TensorFlow 2.0
16. Stanza
Star: 4800, number of submissions: 1514, contributors: 19
The official standard NLP Python library for many human languages
17. TextBlob
Star: 7300, number of submissions: 542, contributors: 24
Simple, Python style, text processing-sentiment analysis, part-of-speech tagging, noun phrase extraction, translation, etc.
18. PyTorch-NLP
Star: 1800, submissions: 442, contributors: 15
Basic utility tools for PyTorch Natural Language Processing (NLP)
19. Textacy
Star: 1500, number of submissions: 1324, contributors: 23
The Python library for performing various natural language processing (NLP) tasks is built on a high-performance spaCy library.
20. Finetune
Star: 626, submissions: 1405, contributors: 13
Finetune is a library that allows users to use the latest pre-trained NLP models to perform various downstream tasks.
21. TextHero
Star: 1900, submissions: 266, contributors: 17
Text preprocessing, presentation and visualization, from zero to proficient.
22. Spark NLP
Star: 1700, submissions: 4363, contributors: 50
Spark NLP is a natural language processing library built on Apache Spark ML.
23. GluonNLP
Star: 2200, submissions: 712, contributors: 72
GluonNLP is a toolkit that simplifies text preprocessing, data set loading and neural model construction to help you speed up your research on natural language processing (NLP).
Computer vision
24. Pillow
Star: 7800, number of submissions: 10799, contributors: 303
Pillow is a very user-friendly branch of PIL. PIL is a Python image library
25. OpenCV
Star: 49600, submissions: 29453, contributors: 1234
Open source computer vision library
26. scikit-image
Star: 4000, number of submissions: 12352, contributors: 403
Image processing with Python
27. Mahotas
Star: 644, number of submissions: 1273, contributors: 25
Mahotas is a library that contains the fastest computer vision algorithms (all algorithms are implemented in C++ to ensure running speed), running against numpy arrays
28. Simple-CV
Star: 2400, submissions: 2625, contributors: 69
SimpleCV is an architecture for open source machine vision, using OpenCV and Python programming languages.
29. GluonCV
Star: 4300, submissions: 774, contributors: 101
GluonCV provides the most advanced (SOTA) deep learning model in computer vision.
30. Torchvision
Star: 7500, submissions: 1286, contributors: 334
The Torchvision package contains popular data sets, model architectures, and image conversion methods commonly used in computer vision.