# REF: https://github.com/llSourcell/predicting_stock_prices/demo.py
import sys
import csv
import numpy as np
from sklearn.svm import SVR
import matplotlib.pyplot as plt
class SVR_Solver:
def __init__(self):
self.dates = []
self.prices = []
def get_data(self, filename):
with open(filename, 'r') as csvfile:
csvFileReader = csv.reader(csvfile)
next(csvFileReader) # skipping column names
for row in csvFileReader:
self.prices.append(float(row[1]))
dates = range(1, len(self.prices)+1)
self.dates = np.reshape(dates, (len(dates), 1))
cutpoint = len(self.prices)/10
self.train_date = self.dates[cutpoint+1:]
self.train_price = self.prices[cutpoint+1:]
self.test_date = self.dates[:cutpoint]
self.test_price = self.prices[:cutpoint]
def training(self, c, g):
self.model = SVR(kernel= 'rbf', C= c, gamma= g)
self.model.fit(self.train_date, self.train_price) # fitting the data points in the models
def tune_parameter(self, c_range, g_range):
best = (None, None, 9999999)
for c in xrange(c_range[0], c_range[1]+1):
for g in xrange(g_range[0], g_range[1]+1):
self.training(2**c, 2**g)
pred = self.predict(self.test_date)
mse = self.MSE(pred, self.prices)
if mse < best[2]:
best = (c, g, mse)
print c, g, mse
print best
def MSE(self, pred, real):
return sum([(p - r) ** 2 for p, r in zip(pred, real)]) / len(pred)
def draw(self):
plt.scatter(self.dates, self.prices, color= 'black', label= 'Data')
plt.plot(self.dates, self.model.predict(self.dates), color= 'red', label= 'RBF model')
plt.xlabel('Date')
plt.ylabel('Price')
plt.title('SVR test for SPY trimmed data (2014, 2015)')
#plt.legend()
plt.show()
def predict(self, x):
return self.model.predict(x)
def main():
solver = SVR_Solver()
solver.get_data(sys.argv[1])
solver.tune_parameter([-5, 2], [-15, 1])
#solver.training(1e3, 0.1)
#solver.draw()
#print solver.predict_price(505)
if __name__ == '__main__':
main()
