.
FRDCSA | internal codebases | PSE

PSE Architecture Diagram: GIF

Jump to: Project Description | Capabilities

Project Description

Now that verber is functioning in terms of generating reasonable plans based on synthesized domains, it makes sense to integrate the output of pse into verber. pse is not a fully functional system, and lacks a precise semantics. It mainly requires unfinished work from corpus - i.e. a corpus of todo items. A basic approach we are looking at is modelling everything within Cyc, and using the HTN planning incorporated with Cyc to develop initial plans which can be refined using verber.

Capabilities

• Inventory management can use freekbs for relational store and world modelling : /var/lib/myfrdcsa/codebases/internal/pse/misc/scheduler/2/domains
• If a message is flagged as a certain type of todo, it can be tagged in the pse or score dbs as completed. There will of course be lots of review to ensure that everything is indeed completed.
• /var/lib/myfrdcsa/codebases/internal/unilang/start audience broker clear corpus cso ELog manager OpenCyc pse unilang-Client
• /var/lib/myfrdcsa/codebases/internal/unilang/start audience broker clear corpus cso ELog OpenCyc pse unilang-Client
• Note that in order to determine whether something is working, we look at a few simple things. We know what the work context is from manager, and we know what the current tasks are from score/pse this new thing. So, just ask them from time till time what they got done. Another thing is if a file that is associated with a task is closed, we can queue a question about that task!.
• Maybe it is easy to spot nonsense situations like the one above. Add that to time management system via pse.
• pse could consider outcome expectation, variance...
• This silly evangelist stuff can actually form the basis of pverber and pse! What a riot!
• fix pse-start-showing-goals-when-idle
• Create a formal model of well-being, and use that to reduce problems to for pse.
• Existing task completion cycles should fuel the direction of the pse.
• Maybe its getting time to get field goal working. Now obviously, fieldgoal depends on pse.
• Insert all sorts of hooks for manager/pse to know the current working context.
• One major criteria is that pse should know that certain entries such as capability requests are generally not as important as other ones.
• There are a few things I could work on, and they are related. I can work on the temporal semantics of pse goals in order to improve our ability to satisfy them. Heck maybe you can infer dependencies on the basis of concept/temporal semantics.
• Still need to convert existing pse.log to SQL.
• score or pse should implement critical path and WBS stuff from Schaum's.
• Work on part for classifying tasks into various logics and keeping track of their progress using score/pse, by recording successful completion of tasks.
• Get task completion operational in pse.
• Event system (which works now) should use pse pop up for reminders
• Take the pse goals and convert them to goal axioms where possible. Use that ontology information from Opt.
• In thinking it over, although I want to get verber going, I don't think it is as critical as getting something that determines what the critical path is, i.e., getting pse/score fully operational, to the point where we can assert that things have been done. I don't know that we need to get it to the point where it classifies new items, but that we can specifically annotate existing ones.
• boss/architect should support most of those metrics in the Schaums system. Could use pse for scheduling?
• Could add a few rules somehow to aid classifications, for instance, pse at the beginning is often not a pse requirement, etc.
• Here is an idea - for every requirement, we can guarantee some axiomitization of that requirement for the purposes of planning. It is through that axiomitization that we can improve pse or whatever that planning system I was working on is called.
• (Here is the todo for unilang/corpus/pse)
• (We want to get pse working, tonight!)
• (Then we implement the part of pse that does dependency guessing)
• (all other systems, like pse, then reference the SQL table via primary key)
• Be sure to find all instances of pse.log
• Okay, I have the pse.log file loaded into SQL along with a script to load it for us.
• Obviously pse should use a database.
• The pse system should have a mode for recognizing dependencies, and this is where it generates a list of estimated pre and post dependencies for each item, (focusing of course on important items).
• Use local identifiers to make it easier to identify a goal in pse. For instance [0],[1], etc.
• Here is an idea how to get pse working. Get it working on small domains.
• unilang/pse should also consider the changing landscape of requirements, when asserting these.
• Just thought of something quite interesting, while working you can tell the system things like "rename PDDL files to PDDL2.2", and it should know to remind you to do that before this task is done. You assert them in various task contexts, or whatever, as part of pse. It is not real world planning, but computing todo planning, which we haven't considered in enough depth.
• Seems obvious to me now, CHILL is used to map sentences to semantic representations. That is how we formalize messages sent to pse and maybe unilang, etc.
• pse should use a command grammer.
• What does pse do about circular dependencies?
• Eventually convert pse to run off of a database, like mysql or postgresql
• What system should manage internet access planning? pse?
• Therefore pse should record who posted the goal.
• Maybe we can use planner-el as the basis for the pse interface.
• States for pse are finished, not finished, stale, etc.
• Should have redirects, like pse, tell audience, search goals
• This would be as a means to disambiguate items. For instance - "create capabilities management system" - is this done? Well, when corpus is done, it will be, but how does pse know that necessarily.
• pse can from time to time quiz the user about whether certain events have taken place.
• We can measure the typing rates and improve people's hacker score - which consists of type rate * LOC * pse rating, etc,etc,et
• In other words, pse ought to affect the world state.
• Here are some different kinds of pse input. Do this when this happens.
• Another thing is we can have a "skills explorer" or something that maps out accomplishments from pse to job-search.
• It would be at least possible, perhaps not desirable, to actually use fweb project pages to display output information - for instance, pse, could display public goals through fweb.
• Note that CELT could be used for audience/Machiavelli/pse, for audience for ACLs, for Machiavelli for who should know what, and for pse for plan information and proof - mixed with CPR (core plan representation) and those kinds of things at teknowledge
• pse should look at: http://reliant.teknowledge.com/CPR2/Reports/CPR-RFC4/Design.html#_Toc435005567
• We can use CELT and Snark with pse to ensure that goals are met.
• Finish classification and get pse basically working.
• This would list the most recent tasks, which would be useful for determining whether they had been completed or not. Maybe that's how pse should work. These tasks will sit there unless accomplished, since pse will index them and record whether they have been accomplished or not, until such time as the user says they are not accomplished or specifies a dependency.
• Should set verber up to do that stuff like "verber -t laundry" to launch a task like doing launch. We need a task hierarchy from pse or something. There are certain tasks that are repeatable and we should usually have those in verber.
• I believe I should get something preliminary going for pse pretty soon - as exported from corpus, for now. So that we can begin to get an agenda in place. From this agenda, and from the interest mapping system, we can start getting activities going.
• numerous ideas on the subject: should have classes be a type hierarchy. Maybe build a grpahical tool to interactivcely classify these. Convert these system over to kbfs::Cache, keeping a copy of current method. For each icodebase and common (agentified), add a class. For instance, rather than classifying into a vague category, you could classy directly to "pse, goal", etc.
• Maybe pse should maintain world model?
• pse or verber should keep track of which goals are currently scheduled by verber, and which have been completed.
• /var/lib/myfrdcsa/codebases/internal/unilang/start pse OpenCyc unilang-Client
• Note we can still use HTN's for pse, i.e. cyc, because we can simply state these goals and have a small set of preconditions that just assure tasks are completed.
• Maybe corpus could handle formalization of everything - from verber and pse entries to?
• RSR can take as input for productivity output from pse/verber.
• If I restale a goal pse should recognize this means to reprioritize it.
• pse should be able to list everything that should be done that morning.
• pse should take note of what I am doing in order to help me along.
• Need to have some kind of Emacs pse interface, with features like add task, show priority queue, etc. These tasks could have contexts.
• With verber looking more complete, a lot of this stuff is more possible. Now we just need to debug pse.
• pse can handle deconfliction.
• Note that this very log file is discontinguous and the recent parts are strangely found here: /home/jasayne/.pse/pse.log
• This protocal can actually be integrated into pse by default!
• Can use pse to schedule things.
• We can also use the event system to schedule what we work on. So, for instance, coordinating verber,pse and Event, we can create a plan for going to the robotics club to resolder the connections to the laptop.
• Have automatic dependency estimation for pse.
• Since pse doesn't have goals added all that much, calling save after every goal is probably a good idea.
• I think this strange tmon.out file is being generated with the pse, print command
• /var/lib/myfrdcsa/codebases/internal/unilang/start pse opencyc unilang-client
• Here are some of the things we need to do next: fix the way pse craps out all the time, get Clairvoyance working and with some basic document management systems, including for instance authorized reports to ensure back to me that various people are learning various things (based on testing), find some way to get those recipes normalized, get shops up and running and taking inventory of everything I have, add this information to verber, get MDR (minimum detection route) planning operational, get new tagsets for AWB/Predator working, fix problem with script determining perl dependencies, make packages of my systems and upload, fix mini-dinstall problem, write tutorial on agentification, agentify or otherwise get command line bugzilla working, fix up manager with sleep learning capabilities, create a sample course for clairvoyance, and run backup. You're such a lazy guy, you know that?