Various fixes, better loading of params from files, first tests

Author: marcharper

analyze_data.py

@@ -1,22 +1,18 @@
-from operator import itemgetter
+import csv
 import sys
 
-def read_top_tables(filename, n):
+def read_data(filename):
     """Read in the n top performing results from a given file"""
     results = []
-    with open(filename) as data:
-        for line in data:
-            results.append(line.split(','))
-
     with open(filename) as data:
         reader = csv.reader(data)
         for line in reader:
-            results.append((float(line[3]), int(line[0]), line[-1]))
-    results.sort(reverse=True, key=itemgetter(0))
-    return results[:n]
+            results.append((float(line[-2]), line[-1]))
+    return results
 
 if __name__ == "__main__":
     data_filename = sys.argv[1]
-    results = read_top_tables(data_filename, 10)
-    for result in results:
-        print(result[0], result[-1])
+    results = read_data(data_filename)
+    results.sort()
+    for result in results[-10:]:
+        print(result[0], result[1])

analyze_weights.py

@@ -15,11 +15,11 @@
 Options:
     -h --help                   Show help
     --generations GENERATIONS   Generations to run the EA [default: 500]
-    --population POPULATION     Starting population size  [default: 10]
+    --population POPULATION     Starting population size  [default: 40]
     --mu MUTATION_RATE          Mutation rate [default: 0.1]
-    --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 5]
+    --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 10]
     --processes PROCESSES       Number of processes to use [default: 1]
-    --output OUTPUT_FILE        File to write data to [default: ann_weights.csv]
+    --output OUTPUT_FILE        File to write data to [default: ann_params.csv]
     --objective OBJECTIVE       Objective function [default: score]
     --repetitions REPETITIONS   Repetitions in objective [default: 100]
     --turns TURNS               Turns in each match [default: 200]
@@ -96,7 +96,7 @@ def mutate_weights(weights, num_features, num_hidden, mutation_rate,
         for i, r in enumerate(randoms):
             if r < mutation_rate:
                 p = 1 + random.uniform(-1, 1) * mutation_distance
-                weights[i] = weights[i] * p
+                weights[i] *= p
         return weights
 
     def mutate(self):
@@ -128,8 +128,8 @@ def __repr__(self):
     @classmethod
     def parse_repr(cls, s):
         elements = list(map(float, s.split(':')))
-        num_features = elements[0]
-        num_hidden = elements[1]
+        num_features = int(elements[0])
+        num_hidden = int(elements[1])
         weights = elements[2:]
         return cls(num_features, num_hidden, weights)
 

ann_evolve.py

@@ -145,7 +145,7 @@ def __init__(self, params_class, params_args, size, objective, output_filename,
         self.params_class = params_class
         self.bottleneck = bottleneck
         self.pool = Pool(processes=processes)
-        self.outputer = Outputer(output_filename)
+        self.outputer = Outputer(output_filename, mode='a')
         self.size = size
         self.objective = objective
         if not bottleneck:
@@ -204,12 +204,16 @@ def evolve(self):
         ## Next Population
         indices_to_keep = [p for (s, p) in results[0: self.bottleneck]]
         self.subset_population(indices_to_keep)
-        # Add variants
+        # Add mutants of the best players
+        best_mutants = [p.copy() for p in self.population]
+        for p in best_mutants:
+            p.mutate()
+            self.population.append(p)
+        # Add random variants
         random_params = [self.params_class(*self.params_args)
                          for _ in range(self.bottleneck // 2)]
         params_to_modify = [params.copy() for params in self.population]
         params_to_modify += random_params
-
         # Crossover
         size_left = self.size - len(params_to_modify)
         params_to_modify = self.crossover(params_to_modify, size_left)
@@ -230,6 +234,23 @@ def run(self, generations):
         pass
 
 
+def load_params(params_class, filename, num):
+    """Load the best num parameters from the given file."""
+    parser = params_class.parse_repr
+    all_params = []
+
+    with open(filename) as datafile:
+        reader = csv.reader(datafile)
+        for line in reader:
+            score, rep = float(line[-2]), line[-1]
+            all_params.append((score, rep))
+    all_params.sort(reverse=True)
+    best_params = []
+    for score, rep in all_params[:num]:
+        best_params.append(parser(rep))
+    return best_params
+
+
 ## This function is only used for PSO.
 
 def score_for(strategy_factory, objective, args=None, opponents=None):

evolve_utils.py

@@ -15,7 +15,7 @@
     --mu MUTATION_RATE          Mutation rate [default: 0.1]
     --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 10]
     --processes PROCESSES       Number of processes to use [default: 1]
-    --output OUTPUT_FILE        File to write data to [default: fsm_tables.csv]
+    --output OUTPUT_FILE        File to write data to [default: fsm_params.csv]
     --objective OBJECTIVE       Objective function [default: score]
     --repetitions REPETITIONS   Repetitions in objective [default: 100]
     --turns TURNS               Turns in each match [default: 200]
@@ -149,15 +149,15 @@ def __repr__(self):
     @classmethod
     def parse_repr(cls, s):
         rows = []
-        lines = rep.split(':')
+        lines = s.split(':')
         initial_state = int(lines[0])
         initial_action = lines[1]
 
         for line in lines[2:]:
             row = line.split('_')
             rows.append(row)
         num_states = len(rows) // 2
-        return cls(num_states, rows, initial_state, initial_action)
+        return cls(num_states, 0.1, rows, initial_state, initial_action)
 
 
 if __name__ == '__main__':

fsm_evolve.py

@@ -15,7 +15,7 @@
     --mu MUTATION_RATE          Mutation rate [default: 0.1]
     --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 10]
     --processes PROCESSES       Number of processes to use [default: 1]
-    --output OUTPUT_FILE        File to write data to [default: fsm_tables.csv]
+    --output OUTPUT_FILE        File to write data to [default: hmm_params.csv]
     --objective OBJECTIVE       Objective function [default: score]
     --repetitions REPETITIONS   Repetitions in objective [default: 100]
     --turns TURNS               Turns in each match [default: 200]
@@ -118,7 +118,8 @@ def random_params(num_states):
             t_C.append(random_vector(num_states))
             t_D.append(random_vector(num_states))
         initial_state = randrange(num_states)
-        initial_action = choice([C, D])
+        # initial_action = choice([C, D])
+        initial_action = C
         return t_C, t_D, initial_state, initial_action
 
     def randomize(self):
@@ -178,7 +179,7 @@ def repr_rows(rows):
         ss = []
         for row in rows:
             ss.append("_".join(list(map(str, row))))
-        return ":".join(ss)
+        return "|".join(ss)
 
     def __repr__(self):
         return "{}:{}:{}:{}:{}".format(
@@ -189,18 +190,23 @@ def __repr__(self):
             self.repr_rows([self.emission_probabilities])
         )
 
-    # @classmethod
-    # def parse_repr(cls, s):
-    #     rows = []
-    #     lines = rep.split(':')
-    #     initial_state = int(lines[0])
-    #     initial_action = lines[1]
-    #
-    #     for line in lines[2:]:
-    #         row = line.split('_')
-    #         rows.append(row)
-    #     num_states = len(rows) // 2
-    #     return cls(num_states, rows, initial_state, initial_action)
+    @classmethod
+    def parse_repr(cls, s):
+        def parse_matrix(sm):
+            rows = []
+            lines = sm.split('|')
+            for line in lines:
+                row = line.split('_')
+                row = list(map(float, row))
+                rows.append(row)
+            return row
+        lines = s.split(':')
+        initial_state = int(lines[0])
+        initial_action = lines[1]
+        t_C = parse_matrix(lines[2])
+        t_D = parse_matrix(lines[3])
+        ps = parse_matrix(lines[4])
+        return cls(t_C, t_D, ps, initial_state, initial_action)
 
 
 if __name__ == '__main__':

hmm_evolve.py

@@ -10,11 +10,11 @@
 Options:
     -h --help                   Show help
     --generations GENERATIONS   Generations to run the EA [default: 500]
-    --population POPULATION     Starting population size  [default: 10]
+    --population POPULATION     Starting population size  [default: 40]
     --mu MUTATION_RATE          Mutation rate [default: 0.1]
-    --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 5]
+    --bottleneck BOTTLENECK     Number of individuals to keep from each generation [default: 10]
     --processes PROCESSES       Number of processes to use [default: 1]
-    --output OUTPUT_FILE        File to write data to [default: lookup_tables.csv]
+    --output OUTPUT_FILE        File to write data to [default: lookup_params.csv]
     --objective OBJECTIVE       Objective function [default: score]
     --repetitions REPETITIONS   Repetitions in objective [default: 100]
     --turns TURNS               Turns in each match [default: 200]
@@ -131,7 +131,8 @@ def __repr__(self):
     @classmethod
     def parse_repr(cls, s):
         elements = s.split(':')
-        plays, op_plays, op_start_plays, initial_actions, pattern = elements
+        plays, op_plays, op_start_plays = list(map(int, elements[:3]))
+        initial_actions, pattern = elements[-2:]
         keys = create_lookup_table_keys(plays, op_plays, op_start_plays)
         table = dict(zip(keys, pattern))
         return cls(plays, op_plays, op_start_plays,

lookup_evolve.py

@@ -15,7 +15,7 @@
 Options:
     -h --help                   Show help
     --generations GENERATIONS   Generations to run the EA [default: 100]
-    --population POPULATION     Starting population size  [default: 100]
+    --population POPULATION     Starting population size  [default: 40]
     --processes PROCESSES       Number of processes to use [default: 1]
     --output OUTPUT_FILE        File to write data to [default: pso_tables.csv]
     --objective OBJECTIVE       Objective function [default: score]

pso_evolve.py

@@ -0,0 +1,21 @@
+
+import unittest
+
+from hmm_evolve import HMMParams
+from ann_evolve import ANNParams
+from fsm_evolve import FSMParams
+from lookup_evolve import LookerUpParams
+from evolve_utils import load_params
+
+class TestRepr(unittest.TestCase):
+
+    def test_repr_load(self):
+        for params_class, filename in [
+            (ANNParams, "ann_params.csv"),
+            (LookerUpParams, "lookup_params.csv"),
+            (FSMParams, "fsm_params.csv"),
+            (HMMParams, "hmm_params.csv")
+        ]:
+            print(filename)
+            params = load_params(params_class, filename, 10)
+