23-07-26

파티클의 전역 최적값이 이전 회차와 동일할 때 점진적으로 가중치의 감소, 다를 때 순간적으로 두배의 관성치를 주는 방식을 추가

README.md

@@ -1,5 +1,4 @@
 [![Python Package Index publish](https://github.com/jung-geun/PSO/actions/workflows/pypi.yml/badge.svg?event=push)](https://github.com/jung-geun/PSO/actions/workflows/pypi.yml)
-<a href="https://colab.research.google.com/github/jung-geun/PSO/blob/master/pso2keras.ipynb" target="_parent"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a>
 
 # PSO 알고리즘 구현 및 새로운 시도
 
@@ -29,7 +28,7 @@ pso 알고리즘을 사용하여 새로운 학습 방법을 찾는중 입니다
 자동으로 conda 환경을 설정하기 위해서는 다음 명령어를 사용합니다
 
 ```shell
-conda env create -f ./conda_env/environment.yaml
+conda env create -f conda_env/environment.yaml
 ```
 
 현재 python 3.9 버전, tensorflow 2.11 버전에서 테스트 되었습니다
@@ -51,6 +50,8 @@ pso_model = Optimizer(...)
 pso_model.fit(...)
 ```
 
+<a href="https://colab.research.google.com/github/jung-geun/PSO/blob/master/example/pso2mnist.ipynb" target="_parent"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a>
+
 # 현재 진행 상황
 
 ## 1. PSO 알고리즘 구현

mnist.py

@@ -106,15 +106,15 @@ loss = [
 pso_mnist = Optimizer(
     model,
     loss=loss[0],
-    n_particles=1000,
+    n_particles=500,
-    c0=0.4,
+    c0=0.3,
-    c1=0.6,
+    c1=0.5,
-    w_min=0.5,
+    w_min=0.4,
-    w_max=0.8,
+    w_max=0.9,
     negative_swarm=0.1,
-    mutation_swarm=0.2,
+    mutation_swarm=0.3,
-    particle_min=-5,
+    particle_min=-4,
-    particle_max=5,
+    particle_max=4,
 )
 
 best_score = pso_mnist.fit(

pso/optimizer.py

@@ -70,6 +70,7 @@ class Optimizer:
         if random_state is not None:
             np.random.set_state(random_state)
 
+        model.compile(loss=loss, optimizer="sgd", metrics=["accuracy"])
         self.model = model  # 모델 구조
         self.loss = loss  # 손실함수
         self.n_particles = n_particles  # 파티클 개수
@@ -203,7 +204,6 @@ class Optimizer:
             (float): 목적 함수 값
         """
         self.model.set_weights(weights)
-        # self.model.compile(loss=self.loss, optimizer="sgd", metrics=["accuracy"])
         score = self.model.evaluate(x, y, verbose=0)[1]
         if score > 0:
             return 1 / (1 + score)
@@ -221,7 +221,7 @@ class Optimizer:
         save_path: str = "./result",
         renewal: str = "acc",
         empirical_balance: bool = False,
-        Dispersion: bool = False,
+        dispersion: bool = False,
         check_point: int = None,
     ):
         """
@@ -234,12 +234,12 @@ class Optimizer:
             save_path : str - ex) "./result",
             renewal : str ex) "acc" or "loss" or "both",
             empirical_balance : bool - True : EBPSO, False : PSO,
-            Dispersion : bool - True : g_best 의 값을 분산시켜 전역해를 찾음, False : g_best 의 값만 사용
+            dispersion : bool - True : g_best 의 값을 분산시켜 전역해를 찾음, False : g_best 의 값만 사용
             check_point : int - 저장할 위치 - None : 저장 안함
         """
         self.save_path = save_path
         self.empirical_balance = empirical_balance
-        self.Dispersion = Dispersion
+        self.Dispersion = dispersion
 
         self.renewal = renewal
         try:
@@ -336,7 +336,7 @@ class Optimizer:
                         f"acc : {max_score:.4f} loss : {min_loss:.4f}"
                     )
 
-                    if Dispersion:
+                    if dispersion:
                         ts = self.c0 + np.random.rand() * (self.c1 - self.c0)
                         g_, g_sh, g_len = self._encode(self.g_best)
                         decrement = (epochs - (epoch) + 1) / epochs

pso/particle.py

@@ -35,6 +35,8 @@ class Particle:
         self.mutation = mutation
         self.best_score = 0
         self.best_weights = init_weights
+        self.before_best = init_weights
+        self.before_w = 0
         
         del i_w_, s_, l_
         del init_weights
@@ -107,7 +109,6 @@ class Particle:
         Returns:
             (float): 점수
         """
-        # self.model.compile(loss=self.loss, optimizer="sgd", metrics=["accuracy"])
         score = self.model.evaluate(x, y, verbose=0, use_multiprocessing=True)
         if renewal == "acc":
             if score[1] > self.best_score:
@@ -138,6 +139,15 @@ class Particle:
         encode_g, g_sh, g_len = self._encode(weights=g_best)
         r0 = np.random.rand()
         r1 = np.random.rand()
+        encode_before, before_sh, before_len = self._encode(weights=self.before_best)
+
+        if (encode_before != encode_g).all():
+            self.before_w = w
+            w = w + (self.before_w)
+        else:
+            self.before_w *= 0.6
+            w = w + self.before_w
+
         if self.negative:
             new_v = (
                 w * encode_v
@@ -161,6 +171,7 @@ class Particle:
         del encode_v, v_sh, v_len
         del encode_p, p_sh, p_len
         del encode_g, g_sh, g_len
+        del encode_before, before_sh, before_len
         del r0, r1
 
     def _update_velocity_w(self, local_rate, global_rate, w, w_p, w_g, g_best):