Who hasn't heard of the battle between FB's PyTorch and Google's TensorFlow? This article is a great article written by Kirill Dubovikov from abroad, quickly revealing the root causes of the conflict and competition between these two frameworks.
The core of the competition is that the two frameworks are too similar. What are the similarities between the two frameworks?
1. Is an open source library for high-performance numerical computing
2. Get support from large technology companies
3. Have a strong and active support community
4. Is based on Python
5. Use diagrams to represent the flow of data and operations
6. Have detailed records
Taking all these factors into consideration, we can say that almost anything created in one framework can be replicated in another at a similar cost. Therefore, the problem lies here.
Which framework should you use? What are the main differences between each community?
At /Data, we continuously investigate the developer community to track and predict future trends in different technical fields. Especially for machine learning, this competition is crucial. The ubiquitous framework (if any) will have a huge impact on the path the machine learning community will take in the coming years.
With this in mind, we asked the developers and they said they were involved in data science (DS) or machine learning (ML) which of the two frameworks they were using, how they used them, and what they did in the professional field what.
TensorFlow won the competition, but is PyTorch played on the same console?
Among the 3,000 developers involved in ML or DS, we see 43% use PyTorch or TensorFlow.
This 43% is not evenly distributed between the two frameworks. TensorFlow is 3.4 times larger than PyTorch. A total of 86% of ML developers and data scientists said they are currently using TensorFlow, while only 11% use PyTorch.
In addition, PyTorch has more than 50% of the community using TensorFlow. On the other hand, only 15% of the TensorFlow community also uses PyTorch. It looks like TensorFlow is a must, but PyTorch is a good choice.
Who is using PyTorch and who is using TensorFlow? What is the most commonly used for each framework?
Here are other things that stand out:
This is decisive. Compared with PyTorch, TensorFlow is being used in production and is likely to be deployed in the cloud, because the backend experience for TensorFlow users is significantly higher (4.8 compared to 3.8 for PyTorch users).
Compared with PyTorch, its community is composed of more professional machine learning developers (28%), software architects (26%) and in-house programmers (58%). This is likely due to Google’s focus on deployment through APIs such as Tensorflow serving, which has become a key motivation for many developers to adopt TensorFlow when trying to push data products into production environments.
On the other hand, PyTorch is used more than TensorFlow for special models in data analysis and business environments (10%). In the PyTorch community, there are more Python developers (ie developers who use Python as the main language) working on web applications (46%).
In addition, the versatility of this Pythonic framework allows researchers to test ideas with little friction, so it is the framework of choice for the most advanced cutting-edge solutions.
Extended reading: The Google I/O conference is over, but this little detail has been ignored by everyone!
At the Google I/O conference in May of this year, in addition to introducing Coral's performance and its hardware development products, Google also demonstrated a mobile app developed by a domestic team based on Coral hardware (Coral Dev Board) at the conference—— Model Play. It is reported that Model Play is an AI model sharing market for global AI developers.
Model Play not only provides a platform for global developers to display and communicate with AI models, it can also be used with the Coral Dev Board with Edge TPU to accelerate ML inference and preview the effect of model operation in real time on mobile phones.