Online gradient descent algorithm for portfolio management

online_gradient_descent.py

import csv
import numpy as np
import matplotlib.pyplot as plt

def read_file(filename):
    lines = []
    csvreader = csv.reader(open(filename))
    for line in csvreader:
        lines.append(line)

    lines.pop(0)
    return np.array(lines)

data = read_file('data.csv')
data = data[:, 1:]
data = data.astype(np.float)
orig_data = np.copy(data)

prev_row = None
for idx, row in enumerate(data):
    if prev_row is not None:
        this_row = np.copy(row)
        data[idx] = row/prev_row
        prev_row = this_row
    else:
        prev_row = row
        
R = data[1:, :]

np.random.seed(0)
X = np.array([np.random.uniform(0, 1) for _ in range(R.shape[1])])
X = X/X.sum()

learning_rate = 0.00005

factors = []
for _ in range(50):

    for row in R:
        factor = np.dot(row.T, X)
        
        if len(factors) != 0:
            factor *= factors[-1]
            
        factors.append(factor)
        
        #gradient descent
        X = X + learning_rate * row
        X = X - np.min(X)
        X = X/np.sum(X)
    
    factors.append(1)

plt.plot(factors)
print("Final %gain by the end of 1000th day is {}".format(factors[-2]))