Releases: opennars/opennars
OpenNARS v3.1.0 (Experimental)
First implementation of Dr. Wang's new design:
- Event Bag removed
- Novel Tasks Bag replaced with Global Buffer
- All input channels (including Narsese Channel) feed tasks into Global Buffer
- Inference results, operator execution feedback and events from internal measures go into Internal Experience Buffer.
- Internal Experience Buffer results go into Global Buffer.
- Task priority enhancement, taking previous belief truth value into account.
Buffer:
- Inherits from Bag but items can only stay for a fixed maximum time duration.
- Additionally spatio-temporal compounding (implications, sequences, parallel conjunctions) happens in the buffer on takeout.
Channel:
- Inherits from Buffer, includes a list of channel-specific operators for channel-specific processing.
Why is this version experimental?
- It allows to perform temporal induction&sequencing on derived events with no common term for the first time, causing additional derived tasks.
- Buffer take out is expensive as it exhaustively performs inference between selected event and all other events in the buffer, making it faster will demand tuning and optimization.
- The duration of global buffer is not tuned as well, making it perform worse on most Q&A tasks.
- it does not include the event bag and operation bag sampling described in "Goal-Directed Procedure Learning", meaning the same procedural inference is now more difficult to carry out, so won't work on most procedure learning tasks so far.
- It emphasizes channels to use a Buffer if possible, while it may turn out that sensory channels need modality-specific data structures.
- The branch will be merged into master once the tuning will be successful and the new design performs comparably well like the previous.
OpenNARS Core v3.0.4 (Stable)
Key new features:
- Global buffer for both input and derived tasks.
- Multistep inference metrics for future control comparisons.
- ^system operator
- NarNode accepts raw string packets
Fixes:
- Operator feedback handling fix.
- Anticipation tolerance improvement.
Metrics:
score sum = 276.8959488784132
score sum with time = 272.02524752717056
avg best time = 51.64193548387097
avg best conf = 0.8932127383174617
For a full release package, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.4
OpenNARS Core v3.0.3 (Stable)
Updated anticipation.
For a full release package, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.3
OpenNARS Core v3.0.2 (Bugfix)
For the full OpenNARS release, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.2
The zip here includes a binary for running OpenNARS in the command line:
https://github.com/opennars/opennars/releases/tag/v3.0.2
Side note: There is also an IRC bot working this binary, runnable with Python 2.7.x: https://gist.github.com/patham9/543fbc6b2f02607331718141521217d1
OpenNARS Core v3.0.1 (Bugfix)
For the full OpenNARS release, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.1
The zip here includes a binary for running OpenNARS in the command line:
https://github.com/opennars/opennars/releases/tag/v3.0.1
Side note: There is also an IRC bot working this binary, runnable with Python 2.7.x: https://gist.github.com/patham9/543fbc6b2f02607331718141521217d1
OpenNARS Core v3.0.0 (Stable)
For the full OpenNARS release, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.0
The zip here includes a binary for running OpenNARS in the command line:
Side note: There is also an IRC bot working this binary, runnable with Python 2.7.x: https://gist.github.com/patham9/543fbc6b2f02607331718141521217d1
OpenNARS Core v3.0.0-RC2
OUTDATED, consider https://github.com/opennars/opennars/releases/tag/v3.0.0 instead.
For the full OpenNARS release, go to https://github.com/opennars/opennars-lab/releases/tag/v3.0.0-RC2
The zip here includes a binary for running OpenNARS in the command line:
OpenNARS Core v3.0.0-RC1
OUTDATED, consider https://github.com/opennars/opennars/releases/tag/v3.0.0 instead.