23-05-24 | 2

pyplot 을 추가하여 loss 와 acc 가 학습할때 어떻게 변화하는지 적용

mnist.ipynb

@@ -12,7 +12,7 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "2023-05-24 12:58:36.275491: E tensorflow/stream_executor/cuda/cuda_blas.cc:2981] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n"
+      "2023-05-24 15:37:52.889357: E tensorflow/stream_executor/cuda/cuda_blas.cc:2981] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n"
      ]
     },
     {
@@ -84,7 +84,25 @@
    "metadata": {
     "tags": []
    },
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "x_train : (28, 28, 1) | y_train : ()\n",
+      "x_test : (28, 28, 1) | y_test : ()\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "init particles position: 100%|██████████| 30/30 [00:00<00:00, 36.65it/s]\n",
+      "init velocities: 100%|██████████| 30/30 [00:00<00:00, 681.12it/s]\n",
+      "Iter 0/20:  13%|#3        | 4/30 [00:04<00:20,  1.28it/s]"
+     ]
+    }
+   ],
    "source": [
     "'''\n",
     "optimizer parameter\n",
@@ -98,21 +116,20 @@
     "pso parameter\n",
     "'''\n",
     "n_particles = 30\n",
-    "maxiter = 50\n",
+    "maxiter = 20\n",
     "# epochs = 1\n",
     "w = 0.8\n",
     "c0 = 0.6\n",
     "c1 = 1.6\n",
     "\n",
-    "def auto_tuning(n_particles=n_particles, maxiter=maxiter, c0=c0, c1=c1, w=w):\n",
+    "\n",
     "x_train, y_train, x_test, y_test = get_data()\n",
     "model = make_model()\n",
     "\n",
     "loss = keras.losses.MeanSquaredError()\n",
-    "    optimizer = keras.optimizers.SGD(lr=lr, momentum=momentun, decay=decay, nesterov=nestrov)\n",
     "\n",
     "\n",
-    "    pso_m = PSO(model=model, loss_method=loss, n_particles=n_particles, x_train=x_train, y_train=y_train)\n",
+    "pso_m = PSO(model=model, loss_method=loss, n_particles=n_particles)\n",
     "# c0 : 지역 최적값 중요도\n",
     "# c1 : 전역 최적값 중요도\n",
     "# w : 관성 (현재 속도를 유지하는 정도)\n",
@@ -139,11 +156,30 @@
     "    json.dump(json_save, f)\n",
     "    f.write(',\\n')\n",
     "\n",
-    "    return model\n",
     "\n",
     "# auto_tuning(n_particles=30, maxiter=1000, c0=0.5, c1=1.5, w=0.75)\n"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1af7569b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "loss_, acc_ = pso_m.all_history()\n",
+    "\n",
+    "plt.subplot(2,1,1)\n",
+    "for layer in all_loss:\n",
+    "    plt.plot(layer)\n",
+    "plt.title('loss history')\n",
+    "\n",
+    "plt.subplot(2,1,2)\n",
+    "for layer in all_acc:\n",
+    "    plt.plot(layer)\n",
+    "plt.title('acc history')"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": 3,
@@ -903,7 +939,7 @@
    ],
    "source": [
     "# print(f\"정답 > {y_test}\")\n",
-    "def get_score(model):\n",
+    "\n",
     "x_train, y_train, x_test, y_test = get_data()\n",
     "\n",
     "predicted_result = model.predict(x_test)\n",
@@ -916,11 +952,11 @@
     "        \n",
     "print(f\"틀린 갯수 > {len(not_correct)}/{len(y_test)}\")\n",
     "\n",
+    "\n",
     "for i in range(3):\n",
     "    plt.imshow(x_test[not_correct[i]].reshape(28,28), cmap='Greys')\n",
     "plt.show() \n",
-    "    \n",
+    "\n"
-    "get_score(auto_tuning(n_particles=30, maxiter=1000, c0=0.5, c1=1.5, w=0.75))"
    ]
   },
   {

mnist.py

@@ -77,7 +77,7 @@ def auto_tuning(n_particles=n_particles, maxiter=maxiter, c0=c0, c1=c1, w=w):
     optimizer = keras.optimizers.SGD(lr=lr, momentum=momentun, decay=decay, nesterov=nestrov)
 
 
-    pso_m = PSO(model=model, loss_method=loss, n_particles=n_particles, x_train=x_train, y_train=y_train)
+    pso_m = PSO(model=model, loss_method=loss, n_particles=n_particles)
     # c0 : 지역 최적값 중요도
     # c1 : 전역 최적값 중요도
     # w : 관성 (현재 속도를 유지하는 정도)
@@ -128,7 +128,7 @@ def get_score(model):
         # plt.imshow(x_test[not_correct[i]].reshape(28,28), cmap='Greys')
     # plt.show() 
     
-get_score(auto_tuning(n_particles=30, maxiter=1000, c0=0.5, c1=1.5, w=0.75))
+get_score(auto_tuning(n_particles=30, maxiter=50, c0=0.5, c1=1.5, w=0.75))
 
 # %%
 

pso_tf.py

@@ -9,7 +9,7 @@ class PSO(object):
     Class implementing PSO algorithm
     """
 
-    def __init__(self, model: keras.models, x_train, y_train, loss_method=keras.losses.MeanSquaredError(), n_particles=5):
+    def __init__(self, model: keras.models, loss_method=keras.losses.MeanSquaredError(), n_particles=5):
         """
         Initialize the key variables.
 
@@ -31,11 +31,9 @@ class PSO(object):
             m = keras.models.model_from_json(self.model_structure)
             m.compile(loss=self.loss_method,
                       optimizer="adam", metrics=["accuracy"])
-            # m.fit(x_train, y_train, epochs=1, batch_size=32, verbose=0) # 결과가 너무 좋지 않아서 처음 초기화 할때 어느정도 위치를 수정
             self.particles_weights[_] = m.get_weights()
             # print(f"shape > {self.particles_weights[_][0]}")
 
-
             # self.particles_weights.append(particle_node)
 
         # print(f"particles_weights > {self.particles_weights}")
@@ -78,7 +76,8 @@ class PSO(object):
             n_particles)]  # 각 파티클의 최적값의 점수
         self.g_best_score = 0                  # 전역 최적값의 점수(초기화 - 무한대)
         self.g_history = []
-        self.all_cost_history = [[] for i in range(n_particles)]
+        self.loss_history = [[] for i in range(n_particles)]
+        self.acc_history = [[] for i in range(n_particles)]
         self.g_best_score_history = []
         self.history = []
 
@@ -224,12 +223,13 @@ class PSO(object):
                     if score[1] > self.g_best_score:
                         self.g_best_score = score[1]
                         self.g_best = self.particles_weights[i].copy()
-                        self.g_history.append(self.g_best)
+                        self.g_history.append(self.g_best.copy())
                         self.g_best_score_history.append(
                             self.g_best_score)
 
                 self.score = score
-                self.all_cost_history[i].append(score)
+                self.loss_history[i].append(score[0])
+                self.acc_history[i].append(score[1])
                 # if self.func(self.particles_weights[i]) < self.func(p_best):
                 # self.p_best[i] = self.particles_weights[i]
                 # if self.
@@ -240,7 +240,7 @@ class PSO(object):
                 # self.g_history.append(self.g_best)
                 # print(f"{i} particle score : {score[0]}")
             print(
-                f"loss avg : {self.score[0]/self.n_particles} | acc avg : {self.score[1]/self.n_particles} | best loss : {self.g_best_score}")
+                f"loss avg : {self.score[0]/self.n_particles} | acc avg : {self.score[1]/self.n_particles} | best score : {self.g_best_score}")
 
             # self.history.append(self.particles_weights.copy())
 
@@ -279,5 +279,5 @@ class PSO(object):
     def global_score_history(self):
         return self.g_best_score_history.copy()
 
-    def all_cost(self):
+    def all_history(self):
-        return self.all_cost_history.copy()
+        return self.loss_history, self.acc_history.copy()