TensorFlow Open Sources TensorFlow Decision Forests (TF-DF). TF-DF is a collection of production-ready algorithms for training, serving, and interpreting decision forest models, including random forests and gradient boosted trees. With the flexibility and composability of TensorFlow and Keras, one can now utilize these models for classification, regression, and ranking tasks.
Decision forests are a class of machine learning algorithms that compete with (and frequently outperform) neural networks in quality and performance, especially when working with tabular data. They’re made up of multiple decision trees, making them simple to use and comprehend – and one can utilize a variety of interpretability tools and approaches that are currently available.
The development and explanation of decision forest models will be easier for beginners. There’s no need to list or pre-process input characteristics (decision forests handle numeric and categorical attributes naturally), establish an architecture, or be concerned about models diverging. After your model has been trained, you can plot it or analyze it with simple statistics.
TensorFlow Decision Forests allows you to train state-of-the-art Decision Forests models in TensorFlow with maximum speed, quality, and lowest effort.
Models with a short inference time will help advanced users (sub-microseconds per example in many cases). In addition, this library provides a lot of flexibility for model exploration and research. Combining neural networks and decision forests, in particular, is simple.
Random forests, gradient-boosted trees, CART, (Lambda)MART, DART, Extra Trees, greedy global growth, oblique trees, one-side-sampling, categorical-set learning, random categorical learning, out-of-bag evaluation, and feature importance, and structural feature importance are among the state-of-the-art Decision Forest training and serving algorithms available.
By making it easy to connect tree-based models with multiple TensorFlow tools, libraries, and platforms like TFX, this library can serve as a bridge to the rich TensorFlow ecosystem.
Users who are new to neural networks can utilize decision forests as an easy method to get started with TensorFlow and then move on to more advanced neural networks.