Changeset 118

Timestamp:

01/30/14 02:10:53 (11 years ago)

Author:

Maciej Komosinski

Message:

added a comment that explains the role of the multiple evaluation phase

File:

• experiments/frams/evolve-speed-vs-gravity/data/scripts/evolve-speed-vs-gravity.script (modified) (1 diff)

experiments/frams/evolve-speed-vs-gravity/data/scripts/evolve-speed-vs-gravity.script

@@ -18 +18 @@
         Simulator.print("%g (x%g) %s" % best.fit % best.popsiz % best.genotype);
         
+        // Now, since we have indeterminism (default.sim used: random initialization of neural states and random placement of creatures),
+        // we cannot trust fitness values that have not been confirmed (averaged) during multiple evaluations.
+        // So we start another phase where we wait until the best genotype is evaluated at least min_evaluations times.
+        // No new genotypes are introduced in this phase.
         ExpParams.stagnation=0; //turn off stagnation detection mechanism
         ExpParams.p_mut=0; //we don't want evolution and new genotypes anymore. We only want to evaluate existing genotypes multiple times