zhaozhengcoder
diff --git a/‎序列预测/PCA去趋势化/dev2/.idea/workspace.xml
Lines changed: 21 additions & 33 deletions b/‎序列预测/PCA去趋势化/dev2/.idea/workspace.xml
Lines changed: 21 additions & 33 deletions
diff --git a/‎序列预测/PCA去趋势化/dev2/main.py
Lines changed: 6 additions & 6 deletions b/‎序列预测/PCA去趋势化/dev2/main.py
Lines changed: 6 additions & 6 deletions
diff --git a/‎序列预测/PCA去趋势化/dev2/main2.py
Lines changed: 3 additions & 3 deletions b/‎序列预测/PCA去趋势化/dev2/main2.py
Lines changed: 3 additions & 3 deletions
diff --git a/‎序列预测/PCA去趋势化/dev2/record.xlsx
11.3 KB b/‎序列预测/PCA去趋势化/dev2/record.xlsx
11.3 KB
@@ -137,8 +137,8 @@ def use_pca(dataset):
     # lstm的hyper-parameter
     hidden_size = 400
     layer_num = 1
-    max_epoch = int(2000 * 0.6)
-    dropout_keep_rate = 1
+    max_epoch = int(2000 * 0.7)
+    dropout_keep_rate = 0.8
 
     # 根据输入数据来决定，train_num训练集大小,input_size输入维度
     train_num, time_step_size, input_size = train_x.shape     # sahpe ：12 * 2 *480
@@ -215,12 +215,12 @@ def print_to_console(i, train_y ,train_y_pred,flag_istrain):
     for i in range(1, max_epoch + 1):
         feed_dict = {x_input: train_x, y_real: train_y, keep_prob: dropout_keep_rate, batch_size: train_num}
         sess.run(train_op, feed_dict=feed_dict)
-        if i % 50 == 0:
+        if i % 200 == 0:
             feed_dict = {x_input: train_x, y_real: train_y, keep_prob: 1.0, batch_size: train_num}
             train_y_pred = sess.run(y_pred, feed_dict=feed_dict)
             #print ("train_y_pred : ",train_y_pred.shape)    #(9,480)
             print_to_console(i,train_y, train_y_pred,1)
-        if i % 50 ==0:
+        if i % 200 ==0:
             feed_dict = {x_input: test_x, y_real: test_y, keep_prob: 1.0, batch_size: test_len}
             test_y_pred = sess.run(y_pred, feed_dict=feed_dict)
             #print ("test_y_pred : ",test_y_pred.shape)      #(3,480)
@@ -230,7 +230,7 @@ def print_to_console(i, train_y ,train_y_pred,flag_istrain):
     y_main = dataset_main[0][time_step:days, :]    # y_main的主成分[2~14] shape 12 * 2 * 480
     y_pre_train_real = y_main[:train_len] + train_y_pred * (y_max - y_min) + y_min   #train_y_pred的shape是 ：(9,480)
     y_raw_train = y_main[:train_len] + train_y * (y_max - y_min) + y_min   # true
-    plt.plot(y_raw_train[0])  #只画第一天
+    plt.plot(y_raw_train[0])         #只画第一天
     plt.plot(y_pre_train_real[0])
     plt.show()
     for i in range(0,train_len):
@@ -239,7 +239,7 @@ def print_to_console(i, train_y ,train_y_pred,flag_istrain):
     y_pre_test_real = y_main[train_len:] + test_y_pred * (y_max - y_min) + y_min
     y_raw_test = y_main[train_len:] + test_y * (y_max - y_min) + y_min   # true
     plt.plot(y_raw_test[0])
-    plt.plot(y_pre_test_real[0])  #只画第一天
+    plt.plot(y_pre_test_real[0])     #只画第一天
     for i in range(0,test_len):
         print("test ",i,"  mre, mae, rmse : ", get_metrics(y_pre_test_real[i],y_raw_test[i]))
     plt.show()
 
@@ -137,8 +137,8 @@ def use_pca(dataset):
     # lstm的hyper-parameter
     hidden_size = 400
     layer_num = 1
-    max_epoch = int(2000 * 0.6)
-    dropout_keep_rate = 1
+    max_epoch = int(2000 * 0.5)
+    dropout_keep_rate = 0.5
 
     # 根据输入数据来决定，train_num训练集大小,input_size输入维度
     train_num, time_step_size, input_size = train_x.shape     # sahpe ：12 * 2 *480
@@ -246,7 +246,7 @@ def print_to_console(i, train_y ,train_y_pred,flag_istrain):
         print("test ",i,"  mre, mae, rmse : ", get_metrics(y_pre_test_real[i],y_raw_test[i]))
     plt.show()
 
-
+### 大功告成 ！！！！
 """
 train mre, mae, rmse :  (-0.029828510559539646, 0.029079688831018635, 0.03781274442185763)
 train mre, mae, rmse :  (-0.014006342486120728, 0.029752569218725298, 0.03840164135038352)