# Run 5 individual experiments experiments.
for k in range(len(Es)):
# Generate an ensemble of n experiments
for j in range(n):
# reset the action probabilities.
lrp.reset_actions()
# Run a single experiment. Terminate if it reaches 10000 iterations.
while (True):
# Define m as the next action predicting the depth of the object.
m = lrp.next_action()
# Defin req as the next detectable object depth.
req = det_obj.request()
# reward if m = req.
resp = env.response(m, req)
if (not resp):
lrp.do_reward(m)
else:
lrp.do_penalty(m)
if (max(lrp.p) > 0.98):
# The best depth counting from 0.
# Break at 98% convergence to a single depth.
bestdepth[np.argmax(lrp.p)] += 1
break
if (j == time_between):
mse.next_env()
det_obj.set_env(mse.env_now())
print("The desired vector is now: " + str(mse.env_now()))
print("The probability vector is: " + str(bestdepth / sum(bestdepth)))
# Generate an ensemble of n experiments
source.goto(-300, depths[k])
receiver.clear()
receiver1.clear()
for i in range(num_actions):
transmission[i].clear()
transmission[i].color("green")
transmission[i].shape("arrow")
transmission[i].shapesize(.5, .5)
transmission[i].penup()
transmission[i].setpos(-300, depths[k])
transmission[i].pendown()
transmission[i].goto(-150, depths[i])
transmission[i].write(mse.env_now()[i])
det_obj.set_env(mse.env_now())
first_uav.set_env(mse1.env_now())
print("The desired vector is now: " + str(mse.env_now()))
# lrp.a = tune.find_optimal_a(lrp, env, det_obj)
# print("Optimized value for a is: " + str(lrp.a))
lrp.a = 0.99999999999999
lrp.b = 0.5
bestdepth = np.zeros(num_actions)
bestdepth1 = np.zeros(num_actions)
current_best = 0
current_best1 = 0
for j in range(n):
# reset the action probabilities.
# lrp.reset_actions()
count = 0
# lrp.b = tune.find_optimal_b(lrp, env, det_obj)