Blogs

The last-command slot in buffer queries allows canonical lisp to see what the last request issued to a module was. Unfortunately, requests are not first-class model elements (can't be stored, retrieved, or even inspected), so !eval!s are required to make any significant sense of them.

There is now a super useful (although still somewhat flaky) tool to probe arbitrary model element parameters (models, productions, chunks, modules, buffers), across time. It comes in two part: runtime probe and viewer.

To use it, you need to enable the GeneralProbe in run configuration and provide it the location of a probe file.

Mad props to those who have braved jACT-R in the past and those who are taking a stab at it now. The past week has seen massive improvements across the board. I've been pumping out releases multiple times a day as issues are brought to my attention and then, usually, fixed.

To all of you, I say thank-you and feel free to keep on pestering me when you think something is wrong. (oh, and.. turn on the automatic update in Eclipse)

So there has always been documentation available for using jACT-R within Eclipse, but I don't think anyone knew about it. So, I've made a few things more obvious. New users will now be greeted with the Welcome screen that includes jACT-R specific hints. Already running? You can see this content via Help - Welcome.

This includes automatic links to the included documentation (opened in your preferred web browser) and two cheatsheets. One cheatsheet is an intro to using Eclipse from the jACT-R perspective. The second is the unit one tutorial.

Wouldn't you know it, once the boss started to play with jACT-R he found issue after issue. Fortunately most of them were either trivial stupid little bugs on my part (mostly related to using the lisp parsers) or mental-model mismatch in Eclipse. This release is not only a bug fix, but also introduces a few new tid-bits:

The latest release has been made. Aside from the usual smattering of bug fixes (nothing too big), there have been many performance improvements between jACT-R and the IDE which will be immediately noticeable in terms of speed and memory usage. Incremental compilation and content assist are now dramatically faster. The memory leaks associated with massive log views now appears to be crushed.

New release is available of core, ide, and actr/s. This covers tons of stuff, there have been quite a few changes, mostly related to compatibility and performance. There is a new framework for dealing with efferent control and it is working well for vocal and I'll be updating motor shortly. I'll be releasing a new augmented tutorial soon-ish.

There is a new release available, this one finally fixes the blocked start-up that could occur. It turns out there is an issue with MINA that would drop a message if the client connected too fast. Irritating, but hacked around until a genuine fix is available.

This release also includes some performance improvements for multiple or embodied models in the neighborhood of 5-15% (all depending). Enjoy.

At the most recent ACT-R workshop, John Laird of SOAR fame, was curious as to the performance of ACT-R. Since ACT-R hasn't been concerned with this (in the past) the numbers weren't available off the top of Dan's head. I had some available, and now Dan and I are working on some common benchmarks, synthetic, but at least they are common.

Here are some early numbers before I dive into realtime issues and database backed models:

Big old batch of bug fixes as I make the mad-dash to the ACT-R workshop. Most of the fixes have to do with concurrency and stability. The intermittent lock-up on start has finally been diagnosed thanks to access to machines with many more cores. It looks as though some long standing motor module issues have finally been resolved allowing commands to be correctly handled when aborted by the model or common reality. Finally, the remote execution tool has been modified so that it can work correctly over VPN connections (the server is now the only one that runs a web service).