CLI tool
tensorflow/decision-forests avatar
tensorflow/decision-forests

TensorFlow Decision Forests: the README's first note tells you to leave, and the tags went 1.11.0, 1.12.0, 1.10.1

A collection of state-of-the-art algorithms for the training, serving and interpretation of Decision Forest models in Keras.

694 stars114 forksPythonApache-2.0

At a glance

What is it?
A Keras-integrated library for training, running and interpreting decision forest models, superseded by the C++ library it is built on. Its quickstart example ends mid-word, the BibTeX block it asks you to paste stops at the doi field, and the newest tag is a lower version number than the one before it.
Who is it for?
Start from YDF, not from here, unless something in your existing code depends on the Keras API specifically. The README's own opening note recommends the migration, and the two libraries convert models in both directions, so the port is a path rather than a rewrite.
Can I use it commercially?
Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 180 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • tensorflow/decision-forests694 ★ · +1
Star history
Repository2026-09-112026-11-15
tensorflow/decision-forests693694

1 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The first note in the document recommends leaving it

The README opens with a recommendation rather than a description. Users are advised to migrate to Yggdrasil Decision Forests, abbreviated YDF, on the stated grounds that it trains the same models as TF-DF but is faster and has more functionality, with a migration guide linked. So the project describes itself as superseded by the library underneath it. That relationship is stated plainly further down as well: TF-DF is powered by Yggdrasil Decision Forest, a library to train and use decision forests in C++, JavaScript, CLI and Go. The consequence for a reader is that the two projects have different shapes. TF-DF is a Keras-integrated Python library for classification, regression and ranking, while YDF is a multi-language library with its own documentation site. The models cross over, since TF-DF models are convertible to YDF models and the reverse conversion is documented too.

Three releases where the newest is the lowest version number

The release history is short and its ordering does not survive a second look. v1.11.0 is dated 2024-10-28. v1.12.0 is dated 2025-03-13. v1.10.1 is dated 2025-03-28, the most recent of the three. So a patch release on the 1.10 line shipped two weeks after the 1.12 release, and the highest tag in the repository is v1.10.1 while the highest version number ever published is v1.12.0. Any tooling that resolves the latest release by comparing version segments will pick the wrong one, and any pin written as 1.12.0 will not include whatever 1.10.1 carried. The repository itself has moved on independently of the tags, with a last push dated 2026-05-19 against a latest release from March 2025, so the two signals have been apart for over a year.

The quickstart example stops in the middle of a comment

The minimal end-to-end example is the part most people copy, and it ends unfinished:

python
import tensorflow_decision_forests as tfdf
import pandas as pd

# Load the dataset in a Pandas dataframe.
train_df = pd.read_csv("project/train.csv")
test_df = pd.read_csv("project/test.csv")

# Convert the dataset into a TensorFlow dataset.
train_ds = tfdf.keras.pd_dataframe_to_tf_dataset(train_df, label="my_label")
test_ds = tfdf.keras.pd_dataframe_to_tf_dataset(test_df, label="my_label")

# Train the model
model = tfdf.keras.RandomForestModel()
model.fit(train_ds)

# Look at the model.
model.summary()

# Evaluate the model.
model.evaluate(test_ds)

# Export to a TensorFlow SavedMo

The last line is a comment cut mid-word, where an export step would have been described. Everything before it is complete and the shape is clear: load two CSV files with pandas, convert them with a Keras helper that takes a label name, fit a RandomForestModel, then summarise and evaluate. The missing step is the one that turns a fitted model into an artifact you can ship.

The BibTeX block you are asked to paste stops at the word doi

The citation section asks anyone using the library in a publication to cite a specific paper, Yggdrasil Decision Forests: A Fast and Extensible Decision Forests Library, and offers three forms of it. The raw form is complete, naming Guillame-Bert et al., KDD 2023, pages 4068 to 4077, with the DOI 10.1145/3580305.3599933. The BibTeX form is not: it fills in authors, title, booktitle, pages, year and url, and then stops at a doi key with no value and no closing brace. Since BibTeX is the format people paste into papers and the raw form is the one they read, the block people actually need is the one that is broken. Note also which paper it is: not a TF-DF paper, but the YDF paper, which is the citation instruction that follows from the project being powered by that library.

Windows is supported through WSL, and the two libraries convert models both ways

Platform support is stated in one sentence: available on Linux and Mac, with Windows users directed to run the library through WSL with Linux. So a Windows user without WSL has no path at all, and the repository offers no native build or alternative. The other platform-shaped fact is the model compatibility story, which runs in both directions. TF-DF models convert to YDF models, and YDF models convert to TF-DF models, with the conversion page documented on the YDF site. That matters more than a normal compatibility note because it means the Keras layer is not a dead end: a model trained inside TF-DF can be exported to the C++, JavaScript, CLI or Go library for serving, and a model trained elsewhere can be brought back in. Nothing in the visible text describes the conversion cost or its fidelity limits.

A Bazel workspace with a pinned toolchain and an example called fake_distributed_training.sh

The root listing describes a Bazel project rather than a plain Python package. It carries a .bazelrc, a .bazelignore, a .bazelversion that pins the toolchain version, and a WORKSPACE file, along with configure/, third_party/, tools/, documentation/ and the package directory itself. The examples folder contains a BUILD file and five runnable scripts, and one of them is named fake_distributed_training.sh, sitting next to distributed_training.py and distributed_hyperparameter_optimization.py. A script whose name declares that it fakes the distributed path is there for testing the orchestration locally, which tells you distributed training is part of the design and that its harness is simulated rather than real. For an end user none of this appears in the install path, which is a single pip command.

One shell command to install, and a documentation folder that carries the troubleshooting

The entire installation instruction is one line:

shell
pip3 install tensorflow_decision_forests --upgrade

The upgrade flag is in the documented command itself, which tells you the library expects to sit alongside whatever version of TensorFlow is already present. Everything harder than that is delegated: the installation page inside the repository's documentation folder is the one pointed at for details, troubleshooting and alternative installation solutions, and the README also links an API reference, guides and tutorials on tensorflow.org, a known-issues page, a changelog and a page of further examples. The contributing side points at a developer manual and a contribution guide, and the contact address for the core team is a role mailbox rather than an individual. The credits list five named developers with their email addresses written in an obfuscated form, an unusual habit for a repository of this size.

Editorial conclusion

Start from YDF, not from here, unless something in your existing code depends on the Keras API specifically. The README's own opening note recommends the migration, and the two libraries convert models in both directions, so the port is a path rather than a rewrite. If you do stay, three things are worth knowing before you file an issue. The newest published tag is not the highest version number, so pin by date rather than by assuming v1.10.1 is older than v1.12.0, and note that the last push to the repository is dated 2026-05-19 while the last release is from March 2025, which means fixes land on the branch without a tag. The quickstart in the README is cut off mid-sentence and the citation block is incomplete, so use the documentation pages on tensorflow.org and the installation page rather than copying from the README. And on Windows, the supported path is WSL with Linux, not a native build.

Frequently asked questions

Is TensorFlow Decision Forests still recommended over Yggdrasil Decision Forests?

No. The README's opening note recommends migrating to YDF, saying it trains the same models but is faster and has more functionality, with a migration guide linked. TF-DF models are also convertible to YDF models and the reverse conversion is documented.

Which platforms does TensorFlow Decision Forests support?

Linux and Mac natively. Windows users are told to use the library through WSL with Linux, and no native Windows build is mentioned.

How do I install TensorFlow Decision Forests?

With pip3 install tensorflow_decision_forests --upgrade. The installation page inside the repository's documentation folder is the one pointed at for details, troubleshooting and alternative installation solutions.

What does TensorFlow Decision Forests do?

It trains, runs and interprets decision forest models such as Random Forests and Gradient Boosted Trees inside TensorFlow, supporting classification, regression and ranking. It is powered by Yggdrasil Decision Forest, a library for C++, JavaScript, CLI and Go.

Which paper should I cite for TensorFlow Decision Forests?

The README asks for the YDF paper: Yggdrasil Decision Forests: A Fast and Extensible Decision Forests Library, KDD 2023, pages 4068-4077, doi 10.1145/3580305.3599933. The BibTeX block provided for it stops at the doi field with no value.

Official sources

  1. Issues
  2. License: Apache-2.0
  3. README
  4. Releases
  5. tensorflow/decision-forests on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/tensorflow-decision-forests.svg)](https://hysenlabs.com/projects/tensorflow-decision-forests)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/tensorflow-decision-forests.svg)](https://hysenlabs.com/projects/tensorflow-decision-forests)

Was this useful?