Tue Sep 8
chapter 2 & 3 : agents, search, and the vacuum world
Terminology from chap 2
- agent, agent function/program, architecture
- environment (or "task environment")
- percept, action
- performance measure
- PEAS : Performance, Environment, Actuators, Sensors
- (deterministic, stochastic), (continuous, discrete), (episodic, sequential)
- simple reflex, model-based reflex, goal-based, utility-based
Ideas from chap 3
- breadth-first
- iterative deepening
- depth-first
- bidirectional
- heuristic
- 8-puzzle (Sterling Blake did his M'boro plan on learning in this sytem, years ago.)
Questions / discussion ?
OK, now for the code ...
Consider a 2D vacuum world, which is part of your homework due Thursday.
Go over the AIMA code and my examples in
code/aima
In class we discussed what a "closure" (google it) looks like in python :
def make_func(data):
def func(x):
return data + x
return func
f1 = make_func(1)
f7 = make_func(7)
print "f1(3) is ", f1(3)
print "f7(3) is ", f7(3)
# The question is: where are the 1 and 7 being stored?
# Once make_func finishes, isn't 'data' out of scope??
# Answer: normally it would be, but since func references it,
# a copy of data is kept around in func's "closure"
# (i.e. environment) ... and there's one of these
# for each of f1 and f7.
python notes
Do read the readme's. Do install, stare at, and play with
their and my source code.
The AIMA files for the vacuum world are utils.py and agents.py .
The python coding in these libraries is more sophisticated
than we've done in the "intro programming" and "algorithms"
courses - now's your chance to learn some new tricks.
Note especially how the code
- passes around functions as arguments
- uses closures to (implicitly) store information between function calls
- (If you don't know what that is, google "closure programming".)
- or uses function attributes, i.e. "def foo(): ... ; foo.color='red';"
What would a good ("intelligent" ?) agent for the 2D vacuum world be?
Can you implement one as a finite state machine (their "model based reflex")
within their API? A good homework problem ... maybe we'll start in class ...
lisp notes
I haven't coded much lisp lately, so I'm rusty.
My code here is from 2011 - if there's interest
I'll brush up on syntax and specifics and discuss Thursday.
Again, look at the readme's for installation notes.
Their two sample agents are defined in agents/agents/vacuum.lisp.
Note that this file shows the glory and pain of lisp;
the (defstructure ...) format won't be found in any lisp manual,
because it's a macro defined by these particular authors,
based on common lisps (defstruct ...) object system, named
after C's structs.
A typical thing to do here is to write your own agent, following
the pattern of theirs ... which means you'll start by looking
at their source code.
The
resources page now has lisp links for those
interested in exploring this language. I particularly
recommend Graham's "ANSI Common Lisp".
I have put lisp notes of my own here, at
jims lisp notes.txt .
But they weren't written as a teaching tool, more
of a "remind me how it works" summary.
for Thursday
So ... see how much you can sink your teeth into this by Thursday,
and come to class ready to share and discuss.
I promise to be more friendly about going over everyone's
code than I was last time. :) (I think I was just short on sleep.)
