init

VitTrainer.py

@@ -44,8 +44,8 @@ class VitTrainer:
             save_strategy="epoch",
             evaluation_strategy="epoch",
             learning_rate=2e-5,
-            per_device_train_batch_size=381,
-            per_device_eval_batch_size=381,
+            per_device_train_batch_size=16,
+            per_device_eval_batch_size=16,
             weight_decay=0.01,
             load_best_model_at_end=True,
             metric_for_best_model="accuracy",
@@ -195,7 +195,8 @@ class VitTrainer:
         print("Data count: ", len(X))
 
         trans = transforms.ToPILImage()
-        X = [trans(torch.tensor([x])) for x in X]
+        #X = [trans(torch.tensor([x])) for x in X]
+        X = [trans(torch.tensor(x)) for x in X]
 
         split_point1 = int(len(X) * 0.9)
         train_X = X[:split_point1]
@@ -221,7 +222,7 @@ if __name__ == "__main__":
     stock_code = "252670"
     vitTrainer = VitTrainer(RESOURCE_PATH)
 
-    train_ds, val_ds = vitTrainer.getData(stock_code, sDate="20200729", eDate="20200731")
+    train_ds, val_ds = vitTrainer.getData(stock_code, sDate="20220701", eDate="20220731")
     vitTrainer.train(train_ds, val_ds, model_path)
 
     """

stock/util/Stock2Vector.py

@@ -195,6 +195,7 @@ class Stock2Vector(HTS):
             result["label"].append(int(label))
 
         return result
+
     def preprocessData(self, result):
         # 분석을 통해서 볼린저밴드 상/하단을 계산한다.
         df = self.buySellChecker.getVectorFeature(result)
@@ -285,12 +286,37 @@ class Stock2Vector(HTS):
         return X, Y
 
     def getVectorData(self, data, type="avg10", VECTOR_SIZE = 32):
-        X, Y = np.zeros((VECTOR_SIZE, VECTOR_SIZE, 4)), np.zeros((32,1))
-        df, minmax_df = self.preprocessData(data)
+        df = self.buySellChecker.getVectorFeature(data)
 
-        X = np.asarray(X)
-        Y = np.asarray(Y, dtype='int64')
-        return X, Y
+        # channel1: avg10, channel2: diff_upper_lower, channel3: abs_avg_2, channel4: abs_avg_3
+        avg10 = df['avg10'].tolist()
+        diff_upper_lower = df['diff_upper_lower'].tolist()
+        rsi = df['rsi'].tolist()
+        abs_avg_3 = df['abs_avg_3'].tolist()
+
+        size = len(avg10)
+        batch_X, batch_Y = [], []
+        X = np.zeros((4, VECTOR_SIZE, VECTOR_SIZE))
+        label = df['label'].tolist()
+        for i in range(VECTOR_SIZE*VECTOR_SIZE-1, size):
+            w, h = 0, 0
+            for j in range(i-VECTOR_SIZE*VECTOR_SIZE+1, i+1):
+                X[0, h, w] = avg10[j]
+                X[1, h, w] = diff_upper_lower[j]
+                X[2, h, w] = abs_avg_3[j]
+                X[3, h, w] = rsi[j]
+                w += 1
+                if w >= VECTOR_SIZE:
+                    w = 0
+                    h += 1
+                    if h >= VECTOR_SIZE:
+                        h = 0
+                        batch_X.append(X)
+            batch_Y.append(label[i])
+
+        batch_X = np.asarray(batch_X)
+        batch_Y = np.asarray(batch_Y, dtype='int64')
+        return batch_X, batch_Y
 
 
     def getDataset3D(self, data, VECTOR_SIZE = 299):