horizon = choice([5, 10, 20])
mix = choice([0, .25, .5, .75, 1])
none_prob = choice([0, None])
thres_coeff = np.random.uniform(0, .5, size=2)
else:
entry_rate = 3
death_rate = .01
n_iters = 5
horizon = 10
mix = .5
none_prob = 0
thres_coeff = 0.5, .01
max_expected_size = entry_rate / death_rate
env = OPTNKidneyExchange(5, .1, 500, seed=12345, initial_size=0)
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 = np.zeros(env.time_length)
#%%
for t in range(env.time_length):
while True:
optsim = OptimalSimulationWithPriors(env,
gamma = 0.99
print(net)
net.opt = torch.optim.Adam(net.parameters(), lr=0.0001)
#%%
for epoch in range(5):
for seed in [0, 1, 2] * 5:
print("Creating environment")
env = OPTNKidneyExchange(entry_rate,
death_rate,
time_length,
seed=seed)
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 = 0
print("Beginning")
for t in range((2 * horizon + 1) * (epoch + 1)):
g_acts = greedy_actions(env, t, n_repeats)
if len(g_acts) == 0:
return 0
values = dict()
for g in g_acts:
snap = snapshot(env, t)
snap.removed_container[t].update(g_acts)
avg_value = np.mean(simulate_value(snap, t, horizon,
gamma, n_iters))
values[g] = avg_value
return values
#%%
env = OPTNKidneyExchange(5, .1, 100, seed = 12345)
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)
#%%
time_limit = env.time_lenth
t = 0
horizon = 10 #int(argv[1])
n_iters = 1 #int(argv[2])
gamma = 1 #float(argv[3])
n_times = 5
rewards = []
matched = set()
if env_type == "abo":
env = ABOKidneyExchange(entry_rate=np.random.randint(5, 10),
death_rate=np.random.choice(
[.1, .09, .08, .07, .06, .05]),
time_length=1000,
seed=s)
elif env_type == "saidman":
env = SaidmanKidneyExchange(entry_rate=np.random.randint(5, 10),
death_rate=np.random.choice(
[.1, .09, .08, .07, .06, .05]),
time_length=1000,
seed=s)
elif env_type == "optn":
env = OPTNKidneyExchange(entry_rate=np.random.randint(5, 10),
death_rate=np.random.choice(
[.1, .09, .08, .07, .06, .05]),
time_length=1000,
seed=s)
else:
env_type = "optn"
s = 1234
env = OPTNKidneyExchange(entry_rate=5,
death_rate=.1,
time_length=1000,
seed=1234)
X, G, E, Y = get_features(env)
x_filename = "X_{}_{}".format(env_type, s)
g_filename = "G_{}_{}".format(env_type, s)
e_filename = "E_{}_{}".format(env_type, s)
a_probs = evaluate_policy(net, env, t)
cycles = list(map(tuple, optimal(env, t, t)["matched_cycles"][t]))
if len(cycles) == 0:
return None
np_probs = np.zeros(len(cycles))
for q, (i, j) in enumerate(cycles):
np_probs[q] = np.mean((a_probs[liv_idx[i]], a_probs[liv_idx[j]]))
selected = np.argmax(np_probs)
if np_probs[selected] >= thres:
return cycles[selected]
else:
return None
#%%
print("Creating environment")
env = OPTNKidneyExchange(entry_rate, death_rate, time_length, seed=12345)
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)
for i in range(1000):
env.removed_container.clear()
print("Solving environment")
rewards = []
actions = []
if platform == "linux":
entry_rate = choice([3, 5, 10])
death_rate = choice([.01, .05, .1])
n_iters = choice([1, 10, 20, 100])
horizon = choice([5, 10, 20])
mix = choice([0, .25, .5, .75, 1])
none_prob = choice([0, None])
else:
entry_rate = 3
death_rate = .01
n_iters = 5
horizon = 3
mix = .5
none_prob = None
env = OPTNKidneyExchange(5, .1, 500, seed=12345, initial_size=0)
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 = np.zeros(env.time_length)
dsdsd
#%%
for t in range(env.time_length):
while True:
optsim = OptimalSimulationWithPriors(env,
thres = 0.3
algo = joblib.load("results/traditional_ml/{}_{}_{}.pkl".format(envtype, algorithm, add))
seed = clock_seed()
args = {'entry_rate': entry_rate,
'death_rate': death_rate,
'time_length': 1001,
'seed': seed}
if envtype == "abo":
env = ABOKidneyExchange(,
elif envtype == "saidman":
env = SaidmanKidneyExchange(,
elif envtype == "optn":
env = OPTNKidneyExchange(**args)
opt = optimal(env)
gre = greedy(env)
g = get_n_matched(gre["matched"], 0, env.time_length)
o = get_n_matched(opt["matched"], 0, env.time_length)
#%%
rewards = np.zeros(env.time_length)
for t in trange(env.time_length):
liv = np.array(env.get_living(t))
A = env.A(t)
has_cycle = np.diag(A @ A) > 0
liv_and_cycle = liv[has_cycle]
else:
frac_ndd = 0
time_length = 1000
t = time()
print("Entry:", entry_rate, "Death", death_rate, "Cycle", max_cycle, "Chain",
max_chain)
env_params = dict(entry_rate=entry_rate,
death_rate=death_rate,
time_length=time_length,
fraction_ndd=frac_ndd)
envname, env = choice([
("OPTN", OPTNKidneyExchange(entry_rate, death_rate, time_length)),
("RSU", SaidmanKidneyExchange(entry_rate, death_rate, time_length)),
("ABO", ABOKidneyExchange(**env_params))
])
print("\tSolving optimal")
opt = optimal(env, max_cycle=max_cycle, max_chain=max_chain)
print("\tSolving greedy")
gre = greedy(env, max_cycle=max_cycle, max_chain=max_chain)
t_diff = time() - t
res = [
envname, env.entry_rate, env.death_rate, env.time_length, frac_ndd,
max_cycle, max_chain, opt["obj"], gre["obj"], gre["obj"] / opt["obj"],
t_diff