Skip to content

Releases: SakanaAI/treequest

v0.3.1

09 Dec 17:04
93b20f3

Choose a tag to compare

Minor fixes of a type error (#19) and a packaging issue (#20)

v0.3.0

25 Nov 07:20
bab11ac

Choose a tag to compare

This release includes a major update on the visualization of the search tree, supporting the file extensions of PNG, PDF, and notably HTML format for interactive visualization (#15, thanks a lot @Yuki-Imajuku!).

See README.md for details.

Feature Highlights

Rich Interactive Interface

HTML format allows us to expand the search tree to see more detailed information about the search.
511910433-78d0c368-46b7-4318-ac08-e51d089c61ef

Easy to Use

We can start using it simply by calling tq.render.

Highly Customizable

We can customize the visualization by using state_formatter.

v0.2.0

18 Oct 00:08
e6865b7

Choose a tag to compare

This release includes a major update on the TreeQuest interface (#14).

Ask-Tell Interface

We added ask-tell interface, where users run ask or ask_batch to get the parent state and action of the next node to expand in the search tree. Using the returned Trial object, users run generate_fn to calculate the score and state of the expanded node, and update the search tree with tell interface.

This new interface enables us to

  1. Perform multiple AB-MCTS sampling steps in parallel by ask_batch.
  2. Run multiple generation processes in parallel and run tell once an evaluation process finishes.

so that now both TreeQuest itself and users can leverage more cpu resource to speed up the tree search.

For the performance for different batch sizes, please see docs/PROFILING.md.

v0.1.2

31 Jul 17:52
57f60f4

Choose a tag to compare

Fixed

  • numpyro import error (#12)

v0.1.1

09 Jul 12:11
f1defc7

Choose a tag to compare

Added

  • Add pre-commit hook (#4)
  • Add PyPi publish GitHub workflow (#5)

Fixed

  • Fixed an error in treequest import when no extra deps are installed (#3, thank you @L0ckR!)

v0.1.0

30 Jun 16:11

Choose a tag to compare

Initial release