23-06-28

단순 업데이트

iris.py

@@ -1,22 +1,24 @@
 import os
+
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
 
 import tensorflow as tf
+
 tf.random.set_seed(777)  # for reproducibility
 
 import numpy as np
+
 np.random.seed(777)
 
+import gc
+
+from pso import Optimizer
 from sklearn.datasets import load_iris
 from sklearn.model_selection import train_test_split
-
 from tensorflow import keras
 from tensorflow.keras import layers
 from tensorflow.keras.models import Sequential
 
-from pso import Optimizer
-
-import gc
 
 def make_model():
     model = Sequential()
@@ -45,7 +47,7 @@ loss = ['categorical_crossentropy']
 pso_iris = Optimizer(
     model,
     loss=loss[0], 
-    n_particles=75, 
+    n_particles=100, 
     c0=0.4, 
     c1=0.8, 
     w_min=0.7,

mnist.py

@@ -59,8 +59,6 @@ def make_model():
 
     return model
 
-# %%
-
 # %%
 model = make_model()
 x_test, y_test = get_data_test()
@@ -73,12 +71,12 @@ if __name__ == "__main__":
             model,
             loss=loss[0], 
             n_particles=75,
-            c0=0.35, 
+            c0=0.3, 
-            c1=0.8, 
+            c1=0.7, 
-            w_min=0.7,
+            w_min=0.6,
-            w_max=1.15,
+            w_max=0.9,
             negative_swarm=0.25,
-            momentun_swarm=0.25,
+            momentun_swarm=0,
             )
 
         best_score = pso_mnist.fit(
@@ -87,13 +85,12 @@ if __name__ == "__main__":
             epochs=200,
             save=True,
             save_path="./result/mnist", 
-            renewal="loss", 
+            renewal="acc", 
             empirical_balance=False,
             Dispersion=False, 
             check_point=25
             )
     except Exception as e:
         print(e)
-
+    finally:
-# pso_mnist.model_save("./result/mnist")
+        gc.collect()
-# pso_mnist.save_info("./result/mnist")

pso/optimizer.py

@@ -12,8 +12,6 @@ from tqdm import tqdm
 
 # import cupy as cp
 
-
-
 gpus = tf.config.experimental.list_physical_devices("GPU")
 if gpus:
     try:
@@ -37,6 +35,8 @@ class Optimizer:
         momentun_swarm: float = 0,
     ):
         """
+        particle swarm optimization
+        
         Args:
             model (keras.models): 모델 구조
             loss (str): 손실함수
@@ -67,7 +67,8 @@ class Optimizer:
             m = keras.models.model_from_json(model.to_json())
             init_weights = m.get_weights()
             w_, sh_, len_ = self._encode(init_weights)
-            w_ = np.random.uniform(-1.5, 1.5, len(w_))
+            w_ = np.random.rand(len(w_)) * 4 - 2 
+            # w_ = np.random.uniform(-1.5, 1.5, len(w_))
             m.set_weights(self._decode(w_, sh_, len_))
             m.compile(loss=self.loss, optimizer="sgd", metrics=["accuracy"])
             self.particles[i] = Particle(m, loss, negative=True if i < negative_swarm * self.n_particles else False, momentun=True if i > self.n_particles * (1 - self.momentun_swarm) else False)
@@ -309,17 +310,23 @@ class Optimizer:
 
                     if renewal == "acc":
                         if score[1] >= self.g_best_score[0]:
-                            self.g_best_score[0] = score[1]
+                            if score[1] > self.g_best_score[0]:
-                            if score[0] <= self.g_best_score[1]:
+                                self.g_best_score[0] = score[1]
-                                self.g_best_score[1] = score[0]
+                                self.g_best = self.particles[i].get_best_weights()
-                            self.g_best = self.particles[i].get_best_weights()
+                            else:
+                                if score[0] < self.g_best_score[1]:
+                                    self.g_best_score[1] = score[0]
+                                    self.g_best = self.particles[i].get_best_weights()
                             epochs_pbar.set_description(f"best {self.g_best_score[0]:.4f} | {self.g_best_score[1]:.4f}")
                     elif renewal == "loss":
                         if score[0] <= self.g_best_score[1]:
-                            self.g_best_score[1] = score[0]
+                            if score[0] < self.g_best_score[1]:
-                            if score[1] >= self.g_best_score[0]:
+                                self.g_best_score[1] = score[0]
-                                self.g_best_score[0] = score[1]
+                                self.g_best = self.particles[i].get_best_weights()
-                            self.g_best = self.particles[i].get_best_weights()
+                            else:
+                                if score[1] > self.g_best_score[0]:
+                                    self.g_best_score[0] = score[1]
+                                    self.g_best = self.particles[i].get_best_weights()
                             epochs_pbar.set_description(f"best {self.g_best_score[0]:.4f} | {self.g_best_score[1]:.4f}")
 
                     if score[0] == None:
@@ -420,6 +427,7 @@ class Optimizer:
             "empirical_balance": self.empirical_balance,
             "Dispersion": self.Dispersion,
             "negative_swarm": self.negative_swarm,
+            "momentun_swarm": self.momentun_swarm,
             "renewal": self.renewal,
         }
 

pso/particle.py

@@ -114,6 +114,8 @@ class Particle:
                 self.best_score = score[1]
                 self.best_weights = self.model.get_weights()
         elif renewal == "loss":
+            if score[0] == 'nan':
+                score[0] = np.inf
             if score[0] < self.best_score:
                 self.best_score = score[0]
                 self.best_weights = self.model.get_weights()

readme.md

@@ -147,7 +147,7 @@ best_score = pso_iris.fit(
 ```
 
 위의 파라미터 기준 2 세대에 94%의 정확도를, 7 세대에 96%, 106 세대에 99.16%의 정확도를 보였습니다
-![iris](./history_plt/iris_relu_acc_200.png)
+![iris](./history_plt/iris_0624_1.png)
 
 3. mnist 문제