refactoring

Author: JordiCorbilla

GOOG_20200705_d0abf0e058c8c64bcb27b9028cecb7fb/Alphabet Inc_hist.png

@@ -15,7 +15,6 @@
 import os
 import secrets
 import pandas as pd
-import tensorflow as tf
 from datetime import datetime
 
 from stock_prediction_class import StockPrediction
@@ -28,27 +27,18 @@
 
 def train_LSTM_network(stock):
     data = StockData(stock)
-
     plotter = Plotter(True, stock.get_project_folder(), data.get_stock_short_name(), data.get_stock_currency(), stock.get_ticker())
-
-    (x_train, y_train), (x_test, y_test), (min_max, test_data) = data.download_transform_to_numpy(TIME_STEPS)
-
-    print(x_test)
+    (x_train, y_train), (x_test, y_test), (training_data, test_data) = data.download_transform_to_numpy(TIME_STEPS)
+    plotter.plot_histogram_data_split(training_data, test_data, stock.get_validation_date())
 
     lstm = LongShortTermMemory(stock.get_project_folder())
     model = lstm.create_model(x_train)
-
-    defined_metrics = [
-        tf.keras.metrics.MeanSquaredError(name='MSE')
-    ]
-
-    callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, mode='min', verbose=1)
-
-    model.compile(optimizer='adam', loss='mean_squared_error', metrics=defined_metrics)
+    model.compile(optimizer='adam', loss='mean_squared_error', metrics=lstm.get_defined_metrics())
     history = model.fit(x_train, y_train, epochs=EPOCHS, batch_size=BATCH_SIZE, validation_data=(x_test, y_test),
-                        callbacks=[callback])
+                        callbacks=[lstm.get_callback()])
     print("saving weights")
     model.save(os.path.join(stock.get_project_folder(), 'model_weights.h5'))
+
     plotter.plot_loss(history)
     plotter.plot_mse(history)
 
@@ -60,7 +50,7 @@ def train_LSTM_network(stock):
 
     print("plotting prediction results")
     test_predictions_baseline = model.predict(x_test)
-    test_predictions_baseline = min_max.inverse_transform(test_predictions_baseline)
+    test_predictions_baseline = data.get_min_max().inverse_transform(test_predictions_baseline)
     test_predictions_baseline = pd.DataFrame(test_predictions_baseline)
     test_predictions_baseline.to_csv(os.path.join(stock.get_project_folder(), 'predictions.csv'))
 

GOOG_20200705_d0abf0e058c8c64bcb27b9028cecb7fb/Alphabet Inc_price.png

@@ -22,6 +22,16 @@ class LongShortTermMemory:
     def __init__(self, project_folder):
         self.project_folder = project_folder
 
+    def get_defined_metrics(self):
+        defined_metrics = [
+            tf.keras.metrics.MeanSquaredError(name='MSE')
+        ]
+        return defined_metrics
+
+    def get_callback(self):
+        callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, mode='min', verbose=1)
+        return callback
+
     def create_model(self, x_train):
         model = Sequential()
         # 1st layer with Dropout regularisation

stock_prediction_deep_learning.py

@@ -26,6 +26,7 @@ class StockData:
     def __init__(self, stock):
         self._stock = stock
         self._sec = yf.Ticker(self._stock.get_ticker())
+        self._min_max = MinMaxScaler(feature_range=(0, 1))
 
     def __data_verification(self, train):
         print('mean:', train.mean(axis=0))
@@ -36,28 +37,27 @@ def __data_verification(self, train):
     def get_stock_short_name(self):
         return self._sec.info['shortName']
 
+    def get_min_max(self):
+        return self._min_max
+
     def get_stock_currency(self):
         return self._sec.info['currency']
 
     def download_transform_to_numpy(self, time_steps):
-        min_max = MinMaxScaler(feature_range=(0, 1))
         end_date = datetime.today()
         print('End Date: ' + end_date.strftime("%Y-%m-%d"))
         data = yf.download([self._stock.get_ticker()], start=self._stock.get_start_date(), end=end_date)[['Close']]
         data = data.reset_index()
-        print(data)
-
-        plotter = Plotter(True, self._stock.get_project_folder(), self._sec.info['shortName'], self._sec.info['currency'], self._stock.get_ticker())
+        #print(data)
 
         training_data = data[data['Date'] < self._stock.get_validation_date()].copy()
         test_data = data[data['Date'] >= self._stock.get_validation_date()].copy()
         training_data = training_data.set_index('Date')
         # Set the data frame index using column Date
         test_data = test_data.set_index('Date')
-        print(test_data)
-        plotter.plot_histogram_data_split(training_data, test_data, self._stock.get_validation_date())
+        #print(test_data)
 
-        train_scaled = min_max.fit_transform(training_data)
+        train_scaled = self._min_max.fit_transform(training_data)
         self.__data_verification(train_scaled)
 
         # Training Data Transformation
@@ -72,7 +72,7 @@ def download_transform_to_numpy(self, time_steps):
 
         total_data = pd.concat((training_data, test_data), axis=0)
         inputs = total_data[len(total_data) - len(test_data) - time_steps:]
-        test_scaled = min_max.fit_transform(inputs)
+        test_scaled = self._min_max.fit_transform(inputs)
 
         # Testing Data Transformation
         x_test = []
@@ -83,16 +83,16 @@ def download_transform_to_numpy(self, time_steps):
 
         x_test, y_test = np.array(x_test), np.array(y_test)
         x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1))
-        return (x_train, y_train), (x_test, y_test), (min_max, test_data)
+        return (x_train, y_train), (x_test, y_test), (training_data, test_data)
 
-    def __daterange(self, start_date, end_date):
+    def __date_range(self, start_date, end_date):
         for n in range(int((end_date - start_date).days)):
             yield start_date + timedelta(n)
 
     def generate_future_data(self, time_steps, min_max, start_date, end_date):
         x_future = []
         y_future = []
-        for single_date in self.__daterange(start_date, end_date):
+        for single_date in self.__date_range(start_date, end_date):
             x_future.append(single_date)
             y_future.append(random.uniform(10, 100))