update mape

data_loader/data_utils.py

@@ -121,9 +121,7 @@ def data_gen(file_path, n_route, train_val_test_ratio, scalar, n_frame, day_slot
             my_matrix = np.array(data_seq)
             scaler = MinMaxScaler()
             scaler.fit(my_matrix)
-            scaler.data_max_
             data_seq = scaler.transform(my_matrix)
-
         #        data_seq=data_seq[~(data_seq==0).all(axis=1), :]
         print(data_seq.shape)
     except FileNotFoundError:
@@ -133,6 +131,16 @@ def data_gen(file_path, n_route, train_val_test_ratio, scalar, n_frame, day_slot
     train_len = int(n_train * length)
     val_len = int(n_val * length)
     test_len = int(n_test * length)
+
+    seq_train = data_seq[:train_len]
+
+    if scalar == 'z_score':
+        x_stats = {'mean': np.mean(seq_train), 'std': np.std(seq_train)}
+    else:
+        x_stats = {'mean': 0, 'std': 1}
+
+    data_seq = z_score(data_seq, x_stats['mean'], x_stats['std'])
+
     seq_train = seq_gen(train_len, data_seq, 0, n_frame, n_route, day_slot)
     seq_val = seq_gen(val_len, data_seq, train_len, n_frame, n_route, day_slot)
     seq_test = seq_gen(test_len, data_seq, train_len + val_len, n_frame, n_route, day_slot)
@@ -149,11 +157,12 @@ def data_gen(file_path, n_route, train_val_test_ratio, scalar, n_frame, day_slot
     # x_stats: dict, the stats for the train dataset, including the value of mean and standard deviation.
 
     # x_train, x_val, x_test: np.array, [sample_size, n_frame, n_route, channel_size].
-    x_train = z_score(seq_train, x_stats['mean'], x_stats['std'])
-    x_val = z_score(seq_val, x_stats['mean'], x_stats['std'])
-    x_test = z_score(seq_test, x_stats['mean'], x_stats['std'])
-
-    x_data = {'train': x_train, 'val': x_val, 'test': x_test}
+    # x_train = z_score(seq_train, x_stats['mean'], x_stats['std'])
+    # x_val = z_score(seq_val, x_stats['mean'], x_stats['std'])
+    # x_test = z_score(seq_test, x_stats['mean'], x_stats['std'])
+    #
+    # x_data = {'train': x_train, 'val': x_val, 'test': x_test}
+    x_data = {'train': seq_train, 'val': seq_val, 'test': seq_test}
     dataset = Dataset(x_data, x_stats)
     return dataset
 

main.py

@@ -22,11 +22,11 @@
 parser.add_argument('--n_route', type=int, default=140)
 parser.add_argument('--n_his', type=int, default=12)
 parser.add_argument('--n_pred', type=int, default=3)
-parser.add_argument('--train_length', type=float, default=15)
-parser.add_argument('--validate_length', type=float, default=1)
-parser.add_argument('--test_length', type=float, default=1)
+parser.add_argument('--train_length', type=float, default=6)
+parser.add_argument('--validate_length', type=float, default=2)
+parser.add_argument('--test_length', type=float, default=2)
 parser.add_argument('--batch_size', type=int, default=50)
-parser.add_argument('--epoch', type=int, default=15)
+parser.add_argument('--epoch', type=int, default=50)
 parser.add_argument('--save', type=int, default=3)
 parser.add_argument('--ks', type=int, default=3)
 parser.add_argument('--kt', type=int, default=3)

utils/math_utils.py

@@ -1,9 +1,10 @@
-
-
 import numpy as np
 import math
+
+import pandas as pd
 from scipy.stats import pearsonr
 
+
 def z_score(x, mean, std):
     '''
     Z-score normalization function: $z = (X - \mu) / \sigma $,
@@ -14,7 +15,12 @@ def z_score(x, mean, std):
     :param std: float, the value of standard deviation.
     :return: np.ndarray, z-score normalized array.
     '''
-    return (x - mean) / std
+    x = (x - mean) / std
+    x[np.where(x == 0)] = np.nan
+    x = pd.DataFrame(x)
+    x = x.fillna(method='ffill', limit=len(x)).fillna(method='bfill', limit=len(x))
+    x = np.asarray(x.values)
+    return x
 
 
 def z_inverse(x, mean, std):
@@ -35,7 +41,7 @@ def MAPE(v, v_):
     :param v_: np.ndarray or int, prediction.
     :return: int, MAPE averages on all elements of input.
     '''
-    return np.mean(np.abs(v_ - v) / (v + 1e-5))
+    return np.mean(np.abs((v_ - v) / v))
 
 
 def RMSE(v, v_):
@@ -57,9 +63,11 @@ def MAE(v, v_):
     '''
     return np.mean(np.abs(v_ - v))
 
+
 def rrse_(y_true, y_pred):
     return np.sqrt(np.sum((y_true - y_pred) ** 2) / np.sum((y_true - np.mean(y_true)) ** 2))
 
+
 def CORR(y_true, y_pred):
     y_true = np.array(y_true, 'float32')
     y_pred = np.array(y_pred, 'float32')
@@ -87,15 +95,14 @@ def evaluation(y, y_, x_stats):
     '''
     dim = len(y_.shape)
 
-    #print(y)
-    #print(y_)
-
+    # print(y)
+    # print(y_)
     if dim == 3:
         # single_step case
         v = z_inverse(y, x_stats['mean'], x_stats['std'])
         v_ = z_inverse(y_, x_stats['mean'], x_stats['std'])
         return np.array([MAPE(v, v_), MAE(v, v_), RMSE(v, v_)])
-        #return np.array([rrse_(v, v_), MAE(v, v_), CORR(v, v_)])
+        # return np.array([rrse_(v, v_), MAE(v, v_), CORR(v, v_)])
     else:
         # multi_step case
         tmp_list = []