Merge branch 'master' of gitlab.doc.ic.ac.uk:yw21218/metarl

autoaug/autoaugment_learners/gen_learner.py → autoaug/autoaugment_learners/GenLearner.py

@@ -171,38 +171,36 @@ class Genetic_learner(AaLearner):
 
     
     def generate_children(self):
-        parent_acc = sorted(self.history, key = lambda x: x[1], reverse=True)[:self.sp_num]
+        parent_acc = sorted(self.history, key = lambda x: x[1], reverse=True)
         parents = [x[0] for x in parent_acc]
         parents_weights = [x[1] for x in parent_acc]
         new_pols = []
         for _ in range(self.num_offspring):
             parent1, parent2 = self.choose_parents(parents, parents_weights)
-            cross_over = random.randrange(1, len(parent2), 1)
+            cross_over = random.randrange(1, int(len(parent2)/2), 1)
+            cross_over2 = random.randrange(int(len(parent2)/2), int(len(parent2)), 1)
             child = parent1[:cross_over]
-            child += parent2[cross_over:]
+            child += parent2[cross_over:int(len(parent2)/2)]
+            child += parent1[int(len(parent2)/2):int(len(parent2)/2)+cross_over2]
+            child += parent2[int(len(parent2)/2)+cross_over2:]
             new_pols.append(child)
         return new_pols
 
     
-    def learn(self, train_dataset, test_dataset, child_network_architecture, iterations = 10):
+    def learn(self, train_dataset, test_dataset, child_network_architecture, iterations = 100):
 
         for idx in range(iterations):
-            print("iteration: ", idx)
-            if len(self.history) < self.sp_num:
+            print("ITERATION: ", idx)
+            if len(self.history) < self.num_offspring:
                 policy = [self.gen_random_subpol()]
             else:
                 policy = self.bin_to_subpol(random.choice(self.generate_children()))
             
-            print("policyu: ", policy)
-
             reward = self._test_autoaugment_policy(policy,
                                                 child_network_architecture,
                                                 train_dataset,
                                                 test_dataset)  
-            print("reward: ", reward)          
-
-            print("new len hsitory: ", len(self.history))
-            print("hsitory: ", self.history)
+            print("reward: ", reward)
 
 
 

ttest_gen.py

@@ -6,7 +6,7 @@ import torchvision
 
 import autoaug.child_networks as cn
 from autoaug.autoaugment_learners.AaLearner import AaLearner
-from autoaug.autoaugment_learners.gen_learner import Genetic_learner
+from autoaug.autoaugment_learners.GenLearner import Genetic_learner
 
 import random
     
@@ -29,14 +29,15 @@ agent = Genetic_learner(
                             learning_rate=0.05,
                             max_epochs=float('inf'),
                             early_stop_num=10,
+                            num_offspring=10
                             )
 
 
 agent.learn(train_dataset,
             test_dataset,
             child_network_architecture=child_network_architecture,
-            iterations=10)
+            iterations=100)
 
-# with open('randomsearch_logs.pkl', 'wb') as file:
-#                 pickle.dump(self.history, file)
-print(agent.history)
\ No newline at end of file
+# with open('genetic_logs.pkl', 'wb') as file:
+#                 pickle.dump(agent.history, file)
+print(sorted(agent.history, key = lambda x: x[1]))
\ No newline at end of file