4b5f6967ad76666a376bd69dcb5aaf966c8facb6,variance_reduction/baseline_comparison_prototype.py,,,#,92

Before Change

        // action = 1 if np.random.uniform() < tfprob else 0
        // action = tfprob
        action = 5 * (tfprob - 0.5)
        print(action)

        xs.append(x)  // observation
        // y = 1 if action == 0 else 0  // a "fake label"
        ys.append(action)

After Change

        // Run the policy network and get an action to take.
        tfscore = sess.run(score, feed_dict={observations: x})
        action = np.random.normal(loc=tfscore[0][:Out],
                                  scale=np.abs(tfscore[0][Out:]))

        xs.append(x)  // observation
        ys.append(action)

Instances