def rollout(env, t_begin, t_end, taken, gamma=0.97):
snap = snapshot(env, t_begin)
snap.populate(t_begin + 1, t_end, seed=clock_seed())
snap.removed_container[t_begin].update(taken)
value = greedy(snap, t_begin + 1, t_end)
matched = get_n_matched(value["matched"], t_begin, t_end)
matched[0] = len(taken)
return disc_mean(matched, gamma)
def rollout(env, t_begin, t_end, taken, gamma):
snap = snapshot(env, t_begin)
snap.populate(t_begin+1, t_end, seed = clock_seed())
snap.removed_container[t_begin].update(taken)
# opt = optimal(snap, t_begin+1, t_end)
# opt_matched = get_n_matched(opt["matched"], t_begin, t_end)
# opt_matched[0] = len(taken)
# opt_value = disc_mean(opt_matched, gamma)
g = greedy(snap, t_begin+1, t_end)
g_matched = get_n_matched(g["matched"], t_begin, t_end)
g_matched[0] = len(taken)
g_value = disc_mean(g_matched, gamma)
r = g_value #- g_value
return r
newseed = str(np.random.randint(1e8))
train = True
disc = 0.1
net = torch.load("results/RNN_50-1-abo_4386504")
#%%
for k in [2]:
print("Creating environment")
env = ABOKidneyExchange(entry_rate, death_rate, time_length, seed=k)
print("Solving environment")
opt = optimal(env)
gre = greedy(env)
o = get_n_matched(opt["matched"], 0, env.time_length)
g = get_n_matched(gre["matched"], 0, env.time_length)
rewards = []
actions = []
t = -1
print("Beginning")
#%%
for t in range(env.time_length):
living = np.array(env.get_living(t))
if len(living) == 1:
continue