Summary
- Fixed unexpected behavior when sampling from trajectory buffer and prioritised trajectory buffer.
- Added ability to sample from multiple buffers (mixed experience replay).
- Fixed some checks and test cases.
What's Changed
- Fix Jumanji environment spec access in examples by @rk1a in #24
- Update README.md by @EdanToledo in #25
- fix: linter for readme by @EdanToledo in #28
- fix: be explicit about squeeze dim in prioritised sampling to avoid flattening (1,1) arrays by @callumtilbury in #27
- feat: add a prioritised item buffer by @EdanToledo in #19
- chore: Type ignore on max_length_time_axis by @callumtilbury in #35
- feat: simple utility to get timestep count by @callumtilbury in #34
- fix: error when vault's rel_dir is a parent directory by @callumtilbury in #32
- jax.tree_map -> jax.tree.map by @garymm in #38
- feat: Mixed Experience Replay 🤝 by @callumtilbury in #30
- prioritised_trajectory_buffer: allow adding max_length_time_axis by @garymm in #40
- Treat warnings as errors by @mickvangelderen in #41
- Pass max_length_time_axis instead of max_size by @mickvangelderen in #43
- Address
sample_sequence_length
greater thanmin_length_time_axis
by @mickvangelderen in #45 - Fix: Overflow error in PER by @EdanToledo in #48
- Fix/core trajectory buffer by @EdanToledo in #50
- fix:prioritised trajectory buffer by @EdanToledo in #51
- fix: update release.yml by @SimonDuToit in #54
- chore: bump version by @SimonDuToit in #55
New Contributors
- @rk1a made their first contribution in #24
- @garymm made their first contribution in #38
- @mickvangelderen made their first contribution in #41
- @SimonDuToit made their first contribution in #54
Full Changelog: v0.1.2...v0.1.3