def load_V_Vectors(cluster_v_bar_epsilon, _name):
global V_vectors
global p
# global v_prog
res = open("show_out" + ".txt", "w")
_p = Problem.Problem(
initial=[
{s: [random.randint(0, na - 1)] for s in range(n)},
np.zeros(d, dtype=ftype),
np.zeros((n, d), dtype=ftype),
],
_mdp=m,
_cluster_error=cluster_v_bar_epsilon,
_epsilon=epsilon_error,
)
"""the main problem is defined as a mdp with two errors: epsilon_error: the used error in hierarchical clustering
using cosine similarity metric. cluster_v_bar_epsilon: the used error for stopping criteria in generating
\mathcal{V} set."""
_v_prog = propagation_V.propagation_V(
m=m,
inequalities=generate_inequalities(m.d),
cluster_v_bar_epsilon=cluster_v_bar_epsilon,
epsilon_error=epsilon_error,
)
_v_prog.initialize_LP()
start1 = time.clock()
tem = _v_prog.convex_hull_search(_p)
stop1 = time.clock()
# cProfile.run('v_prog.convex_hull_search(p)')
print >> res, "time v propagation", stop1 - start1
print >> res, "iteration", tem[1]
res.flush()
_V_vectors = tem[0]
V_vectors = _V_vectors
print "V_vectors", _V_vectors
print "len(V_vectors)", len(_V_vectors)
print >> res, "V_vectors", _V_vectors
res.flush()
print >> res, "len(V_vectors)", len(_V_vectors)
res.flush()
p = _p
# v_prog = _v_prog
name = "param" + _name + "v-propagate.dmp"
pp = pickle.Pickler(open(name, "w"))
pp.dump(p)
# pp.dump(v_prog)
pp.dump(V_vectors)
time_find_V_optimal = []
number_asked_queries = []
error = []
for iteration in range(average_on):
_lambda_rand = interior_easy_points(d)
_r = my_mdp.generate_random_reward_function(n, na, d)
m = my_mdp.make_simulate_mdp(n, na, _lambda_rand, _r)
Uvec = m.policy_iteration()
exact = m.initial_states_distribution().dot(Uvec)
p = Problem.Problem(initial=[{s:[random.randint(0,na-1)] for s in range(n)}, np.zeros(d, dtype=ftype), np.zeros((n,d),dtype=ftype)],
_mdp=m, _cluster_error=0.1, _epsilon=0.00001)
v_prog = propagation_V.propagation_V(m=m, d=d, cluster_v_bar_epsilon = _cluster_v_bar_epsilon, epsilon_error = epsilon_error)
start_time_gather_V_bars = time.time()
V_vectors = v_prog.convex_hull_search(p)
time_gather_V_bars.append(time.time() - start_time_gather_V_bars)
print >> log1, 'V_vectors', V_vectors
print >> log1, "exact", exact
log1.flush()
V = V_bar_search.V_bar_search(_mdp= m, _V_bar=V_vectors, lam_random= m.get_lambda())
start_time_find_V_optimal = time.time()
v_opt = V.v_optimal()
print >> log1, 'v_opt', v_opt
log1.flush()
