Deep Reinforcement Learning for Trading: Using Gradient Ascent to Maximize Sharpe Ratio