optimizer added

rl_agent/utils.py

@@ -0,0 +1,34 @@
+import numpy as np
+import torch
+
+class myOptimizer():
+
+    def __init__(self, lr, mu, mu_square, adaptation_rate, transaction_cost):
+        self.lr = lr
+        self.mu = mu
+        self.mu_square = mu_square
+        self.adaptation_rate = adaptation_rate
+        self.last_gradient = 0.0
+        self.transaction_cost = transaction_cost
+
+    def step(self, grad_n, reward, last_observation):
+
+        numerator = self.mu_square - (self.mu * reward)
+        denominator = torch.sqrt((self.mu_square - (self.mu ** 2)) ** 3)
+
+        gradient = numerator / denominator
+
+        self.mu = self.mu + self.adaptation_rate * (reward - self.mu)
+        self.mu_square = self.mu_square + self.adaptation_rate * ((reward ** 2) - self.mu_square)
+
+        current_grad = (-1.0 * self.transaction_cost * grad_n)
+        previous_grad = (last_observation + self.transaction_cost) * self.last_gradient
+
+        gradient = gradient * (current_grad + previous_grad)
+
+        self.last_gradient = grad_n
+
+        return self.lr * gradient
+
+
+