TensorFlow
TensorFlow is an open-source machine learning framework designed and published by Google. It tracks data flow graphs over time. Nodes in the data flow graphs represent machine learning algorithms. The graph edges represent n-dimension arrays (i.e. tensors) passed between the nodes. It offers a very high-level and abstract approach to organizing low-level numerical programming, with supporting libraries that can allow your software to run without changes on a regular CPU, to take advantage of GPUs for high-speed numerical computation - or to run unmodified on a distributed machine learning cluster. It even supports specialized high-performance tensor learning hardware found only in the Google cloud. Now released under an Apache 2.0 open source license, TensorFlow was originally developed by the Google Brain team and intended for internal Google use. Supported platforms include Linux, macOS, Windows, and Android. TensorFlow models can be run without a traditional computer platform in the Google Cloud Machine Learning Engine.
- Nearly transparent support for machine learning on CPU, GPU, and distributed architectures.
- Models can be saved at checkpoints and loaded into any TensorFlow instance.
- Includes high-performance implementations of several of the most useful learning models, such as Stochastic Forests and Long Short-Term Memory Recurrent Neural Networks.
- Models can be loaded into the Google Cloud Machine Learning Engine and trained using Google's tensor learning clusters with bleeding-edge numerical performance.
- State of the art machine learning library.
- High performance, matching the best in the industry.
- Packages are available that will let you easily program voice recognition, machine translation, video tagging, and other advanced artificial intelligence tasks.
- Unique approach allows monitoring the training progress of your models and tracking several metrics.
- Great community support.
- The only GPUs supported are Nvidia GPUs.
- Some machine learning packages support more types of models out of the box.
- The only fully supported programming language is Python.
- Some gaps in documentation.
- Sometimes updates break backward compatibility.
TensorFlow is one of the most advanced open source machine learning libraries available today. The learning curve is a bit steep, and the software documentation tracks a fast-moving target, with many incompatible changes made to both the API and the major supporting programming language, Python. The source code examples provided by Google don't always work with the latest software versions. TensorFlow models are accessible with most common programming languages that you might want to use for numerical programming, but the library only fully supports Python for training new models. Performance-wise, TensorFlow ranks with the very best machine learning libraries.
Alternatives to TensorFlow
Torch
Azure Machine Learning
