def __init__(self, stock_to_predict, days_of_prediction = 10, days_of_training = 450):
self.number_of_days_before = 8
self.days_of_prediction = days_of_prediction
self.downloader = StockDownloader()
stock_training_data = self.downloader.download_stock(stock_to_predict, days_of_training, days_of_prediction)
self.stock_prediction_data = self.downloader.download_stock(stock_to_predict, days_of_prediction)
self.starting_price = self.stock_prediction_data[0]
self.dataset = StockSupervisedDataSet(self.number_of_days_before, stock_training_data)
self.network = buildNetwork(self.dataset.indim, 10, self.dataset.outdim, recurrent=True)
t = BackpropTrainer(self.network, learningrate = 0.00005, momentum=0., verbose = True)
t.trainOnDataset(self.dataset, 200)
t.testOnData(verbose= True)
self.starting_prices = self.dataset['input'][-1]
class StockPredicter(object):
def __init__(self, stock_to_predict, days_of_prediction = 10, days_of_training = 450):
self.number_of_days_before = 8
self.days_of_prediction = days_of_prediction
self.downloader = StockDownloader()
stock_training_data = self.downloader.download_stock(stock_to_predict, days_of_training, days_of_prediction)
self.stock_prediction_data = self.downloader.download_stock(stock_to_predict, days_of_prediction)
self.starting_price = self.stock_prediction_data[0]
self.dataset = StockSupervisedDataSet(self.number_of_days_before, stock_training_data)
self.network = buildNetwork(self.dataset.indim, 10, self.dataset.outdim, recurrent=True)
t = BackpropTrainer(self.network, learningrate = 0.00005, momentum=0., verbose = True)
t.trainOnDataset(self.dataset, 200)
t.testOnData(verbose= True)
self.starting_prices = self.dataset['input'][-1]
def predict_with_starting_price_only(self):
#Get predicted price and reverse predicted price
prices = [self.starting_price]
augmentations = list(self.starting_prices)
for i in range(len(self.stock_prediction_data) - 1):
predicted_augmentation = float(self.network.activate(augmentations[-self.number_of_days_before:]))
augmentations.append(predicted_augmentation)
new_price = prices[-1] * (1 + predicted_augmentation)
prices.append(new_price)
moving_avg = self.movingaverage(np.array(prices), 5)
#Plot real prices
figure()
plot(self.stock_prediction_data, color = "black")
#Plot predicted
plot(moving_avg, "green")
show()
def movingaverage(self, interval, window_size):
window = np.ones(int(window_size))/float(window_size)
avg = np.convolve(interval, window, 'same')
half_window = ceil(window_size/2)
avg[:half_window] = interval[:half_window]
avg[-half_window:] = interval[-half_window:]
return avg
def predict_one_day_ahead(self):
figure()
#Get predicted price and reverse predicted price
reverse = []
predicted = []
price = self.starting_price
reverse_price = self.starting_price
for i in range(len(self.dataset)):
if i >= self.number_of_days_before:
predicted_augmentation = self.network.activate(self.dataset['input'][i-self.number_of_days_before])
price = price * (1 + predicted_augmentation)
reverse_price = reverse_price * (1 - predicted_augmentation)
predicted.append(price)
reverse.append(reverse_price)
#Get real price
price = self.starting_price
real = []
for i, target in enumerate(self.dataset['target']):
if i >= self.number_of_days_before:
target = self.dataset['target'][i-self.number_of_days_before]
price = price * (1+target)
real.append(price)
plot(real, color = "black")
plot(predicted, color = "red")
plot(reverse, "blue")
show()
