def root_mean_square_error(y_true, y_pred, n=0):
"""Root mean squared error.
Args:
y_true -- truth value
y_pred -- Predicted value
n -- number of periods
"""
if n == 0:
# This is directly from the sklearn metric
mse = sk_mse(y_true, y_pred)
result = math.sqrt(mse)
return result
else:
# This is directly from the sklearn metric
mse = sk_mse(y_true, y_pred, n)
result = math.sqrt(mse)
return result
def mean_squared_error(y_true, y_pred):
"""Root mean squared error (Overloaded from sklearn).
Args:
y_true -- truth value
y_pred -- Predicted value
n -- number of periods
"""
mse = sk_mse(y_true, y_pred)
return mse
def predict(self,
stock_prices,
stock_processor,
window_len=5,
step_size=1,
plot=True):
'''
Predict stock prices for the given day given last day
Runs for the entire length of stock given
Prints RMSE
::param stock_prices np.array - prices normalised and stationary
::param stock_processor StockProcessor instance (for denorming etc)
::param window_len int number of days to predict
::param step_size int step size of window
'''
preds = []
logging.debug(stock_prices[0])
stock_windows = rolling_window(stock_prices, window_len, step_size)
for idx, price_window in enumerate(stock_windows[:, ]):
if idx == 0:
preds.append(np.zeros(5))
else:
preds.append(stock_windows[idx - 1])
#Check shapes
preds = np.array(preds)
logging.debug(preds[0:3])
try:
assert preds.shape == stock_windows.shape
except AssertionError:
logging.exception('Shape error: %s, %s', preds.shape,
stock_windows.shape)
'''
#De norm and re-trend to get real prices
preds = stock_processor.de_normalise(preds)
stock_prices = stock_processor.de_normalise(stock_prices)
preds = stock_processor.re_trend(preds)
stock_prices = stock_processor.re_trend(stock_prices)
'''
#Check error
rmse = sqrt(sk_mse(stock_windows, preds))
logging.info('Previous Window Method RMSE:%s', rmse)
if plot:
plt.plot(stock_prices, label='Stock')
plt.plot(preds, label='Preds')
leg = plt.legend()
plt.show()
def print_accuracy():
# Cosine mse
y_true = [None, None]
y_true[0] = [0.0 for i in range(len(_cosine_pred[0]))]
y_true[1] = [1.0 for i in range(len(_cosine_pred[1]))]
cosine_dis_mse = sk_mse(_cosine_pred[0], y_true[0])
cosine_sim_mse = sk_mse(_cosine_pred[1], y_true[1])
cosine_tot_mse = sk_mse(_cosine_pred[0] + _cosine_pred[1], y_true[0] + y_true[1])
simnet_dis_mse = sk_mse(_simnet_pred[0], y_true[0])
simnet_sim_mse = sk_mse(_simnet_pred[1], y_true[1])
simnet_tot_mse = sk_mse(_simnet_pred[0] + _simnet_pred[1], y_true[0] + y_true[1])
print("Disimilarity error: cosine={}, simnet={}", cosine_dis_mse, simnet_dis_mse)
print("Similarity error: cosine={}, simnet={}", cosine_sim_mse, simnet_sim_mse)
print("Total error: cosine={}, simnet={}", cosine_tot_mse, simnet_tot_mse)
def predict(self, stock_prices, stock_processor, plot=True):
'''
Predict stock prices for the given day given last day
Runs for the entire length of stock given
Prints RMSE
::param stock_prices np.array - prices normalised and stationary
::param stock_processor StockProcessor instance (for denorming etc)
::param time_period int number of days to predict
'''
preds = []
logging.debug(stock_prices[0])
for idx, price in enumerate(stock_prices[:]):
if idx == 0:
preds.append(0.0)
else:
preds.append(stock_prices[idx - 1])
#Check shapes
preds = np.array(preds)
try:
assert preds.shape[0] == stock_prices.shape[0]
except AssertionError:
logging.exception('Shape error: %s, %s', preds.shape[0],
stock_prices.shape[0])
#De norm and re-trend to get real prices
preds = stock_processor.de_normalise(preds)
stock_prices = stock_processor.de_normalise(stock_prices)
preds = stock_processor.re_trend(preds)
stock_prices = stock_processor.re_trend(stock_prices)
#Check error
rmse = sqrt(sk_mse(stock_prices, preds))
logging.info('Previous Method RMSE:%s', rmse)
if plot:
plt.plot(stock_prices, label='Stock')
plt.plot(preds, label='Preds')
leg = plt.legend()
plt.show()
def mse(self, predictions, targets):
"""
Calculate Mean Squered Error
"""
mse = sk_mse(predictions, targets)
return mse
pred_outputs = sp.de_normalise(np.array(pred_outputs))
tst_outputs = sp.de_normalise(np.array(tst_outputs))
silver_prices = silver_prices.reshape(gdelt_feats.shape[0])
silver_prices = sp.de_normalise(silver_prices[train_outputs.shape[0]:])
logging.info(silver_prices.shape)
#Not sure this is actually working for test data shifted - esp windows
'''
pred_outputs = sp.re_trend(pred_outputs)
tst_outputs = sp.re_trend(tst_outputs)
prices = sp.re_trend(prices)
logging.debug('%s, %s', prices[0:3], prices.shape)
'''
rmse = sqrt(sk_mse(tst_outputs[:-1], pred_outputs[:-1]))
logging.info('RMS: %s', rmse)
'''
plt.plot(prices, label='Stock')
i = 0
#Only plot a few of them
for idx in range(0, pred_outputs.shape[0], WINDOW_SIZE):
#Plot individual windows - in the right place
X = [idx+i for i in range(WINDOW_SIZE)]
logging.debug([X, pred_outputs[idx]])
i+=1
#if i > 10:
# break
plt.plot(X, pred_outputs[idx], label='Preds')
leg = plt.legend()
plt.show()
