meneguzzi / monitoring Goto Github PK

The new heuristic planner fixes a lot of small bugs, but uses certain different structures and Python 3.x, which may be a problem during merge.

Remember:

• sys.maxint is equivalent to float("inf")
• tuple goal_state is broken into positive_goals and negative_goals

Method monitoring.monitor.evaluate_sensor_on_traces reports the exact same number of positive and negative results for different expressions in domain examples/simple/simple.pddl:

• (-((q) ^ (p))) vs ((-(q)) ^ (-(p)))
• (q [2] r) vs (r [1] q)

In file gp.py lines 43 and 46 two classes are instantiated with parameters regarding maxDepth and minDepth. The first class is for generating nodes and the other is for performing mutation/crossover operations. The problem is that the depth limits are different from one another. Is there a reason for that?

Here are the pieces of code:

ng = NodeGenerator(terms, 1, 5, 2, 5)
gpo = GPOps(terms, 1, 5, 1, 4)

See that the last two parameters (maxDepth, minDepth) are different. And then you send these objects to the Population:

pop = Population(popSize, ng, reproducePercent, mutatePercent, crossOverPercent, gpo, sp.parse_sensor(modelSensor), traces, self.ms)

Which means that you perform mutation/crossover operations under different limits of maxDepth and minDepth than the limits you use for generating new nodes in the population. Is there an implementation reason for that?