def plot_trajectory_result_necessary_data(fname, accs_at_least=[0.5]):
results = result_dict_from_file(fname)
fun = lambda x: np.mean(x[:, 1])
Ds, Ns, avg_accept_est = gen_sparse_2d_array_from_dict(results, fun)
plt.figure()
for acc_at_least in accs_at_least:
N_at_least = np.zeros(len(Ds))
for i, D in enumerate(Ds):
w = np.where(avg_accept_est[i, :] > acc_at_least)[0]
if len(w) > 0:
N_at_least[i] = np.min(Ns[w])
logger.info("%.2f acc. for D=%d at N=%d" %
(acc_at_least, D, N_at_least[i]))
else:
logger.info("Did not reach %.2f acc. for D=%d" %
(acc_at_least, D))
plt.plot(Ds, N_at_least)
plt.yscale('log')
# plt.xscale('log')
plt.legend(["%.2f acc." % acc_at_least for acc_at_least in accs_at_least],
loc="lower right")
plt.grid(True)
fname_base = fname.split(".")[-2]
plt.savefig(fname_base + "_data_needs_kmc.eps", axis_inches='tight')
def plot_banana_result_mean_N_D(results, D, stat_idx, normalise_by_time=False,
**kwargs):
fun = lambda x:None
Ds, Ns, _ = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 0])
_, _, time_taken_set_up = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 1])
_, _, time_taken_sampling = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, stat_idx])
_, _, avg = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, stat_idx], 25)
_, _, lower = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, stat_idx], 75)
_, _, upper = gen_sparse_2d_array_from_dict(results, fun)
D_ind = Ds == D
time_total = time_taken_sampling[:,D_ind]# + time_taken_set_up[:,D_ind]
time_total = time_total.ravel()
print_table(time_total, "Time total")
normaliser = time_total if normalise_by_time else 1.0
avg = avg[:, D_ind].ravel()/normaliser
err = np.array([np.abs(avg-lower[:, D_ind].ravel()/normaliser),
np.abs(avg-upper[:, D_ind].ravel()/normaliser)])
plt.plot(Ns, avg, kwargs['color'])
plt.errorbar(Ns, avg, err, color=kwargs['color'])
plt.grid(True)
try:
plt.title(kwargs['title'])
except KeyError:
pass
try:
plt.ylim(kwargs['ylim'])
except KeyError:
pass
try:
plt.xlim(kwargs['xlim'])
except KeyError:
pass
try:
plt.xlabel(kwargs['xlabel'])
except KeyError:
pass
def plot_banana_result_mean_D(results, D, stat_idx, normalise_by_time=False,
**kwargs):
fun = lambda x:None
Ds, _ = gen_sparse_1d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:,0])
_, time_taken_set_up = gen_sparse_1d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:,1])
_, time_taken_sampling = gen_sparse_1d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:,stat_idx])
_, avg = gen_sparse_1d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, stat_idx], 25)
_, lower = gen_sparse_1d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, stat_idx], 75)
_, upper = gen_sparse_1d_array_from_dict(results, fun)
D_ind = Ds == D
time_total = time_taken_sampling[D_ind] + time_taken_set_up[D_ind]
time_total = time_total.ravel()
print_table(time_total, "Time total")
normaliser = time_total if normalise_by_time else 1.0
avg = avg[D_ind].ravel()/normaliser
xlim=plt.xlim()
plt.plot(xlim, [avg,avg], color=kwargs['color'])
plt.plot(xlim, [lower,lower], '--', color=kwargs['color'])
plt.plot(xlim, [upper,upper], '--', color=kwargs['color'])
plt.grid(True)
try:
plt.title(kwargs['title'])
except KeyError:
pass
try:
plt.ylim(kwargs['ylim'])
except KeyError:
pass
try:
plt.xlabel(kwargs['xlabel'])
except KeyError:
pass
def plot_fixed_D(results, results_hmc, D):
fun = lambda x: None
_, _ = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: np.mean(x[:, 0])
_, time_taken_set_up_hmc = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: np.mean(x[:, 1])
_, time_taken_sampling_hmc = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: np.mean(x[:, 2])
_, accept_hmc = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: np.mean(x[:, 3])
_, avg_quantile_error_hmc = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: np.mean(x[:, 4])
_, avg_ess_hmc = gen_sparse_1d_array_from_dict(results_hmc, fun)
fun = lambda x: None
Ds, Ns, _ = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 0])
_, _, time_taken_set_up = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 1])
_, _, time_taken_sampling = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 2])
_, _, accept = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 2], 25)
_, _, accept_lower25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 2], 75)
_, _, accept_upper75 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 2], 5)
_, _, accept_lower5 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 2], 95)
_, _, accept_upper95 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 3])
_, _, avg_quantile_error = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 3], 25)
_, _, avg_quantile_error_lower25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 3], 75)
_, _, avg_quantile_error_upper75 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 3], 5)
_, _, avg_quantile_error_lower5 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 3], 95)
_, _, avg_quantile_error_upper95 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 4])
_, _, avg_ess = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 4], 25)
_, _, avg_ess_lower25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 4], 75)
_, _, avg_ess_upper75 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 4], 5)
_, _, avg_ess_lower5 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 4], 95)
_, _, avg_ess_upper95 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 5])
_, _, avg_norm_of_mean = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 5], 25)
_, _, avg_norm_of_mean_lower25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 5], 75)
_, _, avg_norm_of_mean_upper75 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 5], 5)
_, _, avg_norm_of_mean_lower5 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 5], 95)
_, _, avg_norm_of_mean_upper95 = gen_sparse_2d_array_from_dict(results, fun)
D_ind = Ds == D
time_total = time_taken_sampling[:,D_ind] + time_taken_set_up[:,D_ind]
time_total_hmc = time_taken_sampling_hmc[D_ind] + time_taken_set_up_hmc[D_ind]
plt.figure()
plt.plot([Ns.min(), Ns.max()], [accept_hmc[D_ind], accept_hmc[D_ind]], 'r')
plt.plot(Ns, accept[:, D_ind], 'b')
plt.plot(Ns, accept_lower25[:, D_ind], 'b-.')
plt.plot(Ns, accept_upper75[:, D_ind], 'b-.')
plt.fill_between(Ns,
accept_lower5[:, D_ind].ravel(),
accept_upper95[:, D_ind].ravel(),
color="grey", alpha=.5)
plt.title("Avg. acc. prob.")
plt.ylim([0,1.1])
plt.grid(True)
plt.xlabel(r"$n$")
plt.figure()
plt.plot([Ns.min(), Ns.max()], [avg_quantile_error_hmc[D_ind], avg_quantile_error_hmc[D_ind]], 'r')
plt.plot(Ns, avg_quantile_error[:, D_ind])
plt.plot(Ns, avg_quantile_error_lower25[:, D_ind], 'b-.')
plt.plot(Ns, avg_quantile_error_upper75[:, D_ind], 'b-.')
plt.fill_between(Ns,
avg_quantile_error_lower5[:, D_ind].ravel(),
avg_quantile_error_upper95[:, D_ind].ravel(),
color="grey", alpha=.5)
plt.title("Avg. quantile error")
plt.grid(True)
plt.xlabel(r"$n$")
plt.figure()
# plt.plot([Ns.min(), Ns.max()], np.array([avg_ess_hmc[D_ind], avg_ess_hmc[D_ind]])/time_total_hmc, 'r')
plt.plot(Ns, avg_ess[:, D_ind]/time_total, 'b')
plt.plot(Ns, avg_ess_lower25[:, D_ind]/time_total, 'b-.')
plt.plot(Ns, avg_ess_upper75[:, D_ind]/time_total, 'b-.')
plt.fill_between(Ns,
(avg_ess_lower5[:, D_ind]/time_total).ravel(),
(avg_ess_upper95[:, D_ind]/time_total).ravel(),
color="grey", alpha=.5)
plt.title("Avg. ESS/s")
plt.grid(True)
plt.xlabel(r"$n$")
plt.figure()
# plt.plot([Ns.min(), Ns.max()], [avg_quantile_error_hmc[D_ind], avg_quantile_error_hmc[D_ind]], 'r')
plt.plot(Ns, avg_norm_of_mean[:, D_ind])
plt.plot(Ns, avg_norm_of_mean_lower25[:, D_ind], 'b-.')
plt.plot(Ns, avg_norm_of_mean_upper75[:, D_ind], 'b-.')
plt.fill_between(Ns,
avg_norm_of_mean_lower5[:, D_ind].ravel(),
avg_norm_of_mean_upper95[:, D_ind].ravel(),
color="grey", alpha=.5)
plt.title(r"Avg. $\Vert \mathbb E \mathbf{x} \Vert$")
plt.grid(True)
plt.xlabel(r"$n$")
plt.show()
# normalise Hamiltonians
Hs -= Hs.mean()
Hs_est -= Hs_est.mean()
plt.figure()
plot_array(Xs_q, Ys_q, np.exp(G))
plot_2d_trajectory(Qs)
plt.title("HMC")
plt.gca().xaxis.set_visible(False)
plt.gca().yaxis.set_visible(False)
plt.savefig(fname_base + "_hmc.eps", axis_inches="tight")
plt.figure()
plot_array(Xs_q, Ys_q, np.exp(G_est))
plt.plot(Z[:, 0], Z[:, 1], 'bx')
plot_2d_trajectory(Qs_est)
plt.title("KMC")
plt.gca().xaxis.set_visible(False)
plt.gca().yaxis.set_visible(False)
plt.savefig(fname_base + "_kmc.eps", axis_inches="tight")
plt.figure()
plt.title("Momentum")
plot_array(Xs_p, Ys_p, np.exp(M))
plot_2d_trajectory(Ps)
plt.gca().xaxis.set_visible(False)
plt.gca().yaxis.set_visible(False)
plt.savefig(fname_base + "_momentum_hmc.eps", axis_inches="tight")
plt.figure()
def plot_trajectory_result_mean_fixed_N(fname, N):
results = result_dict_from_file(fname)
# acc_mean, acc_est_mean, vol, vol_est, steps_taken
fun = lambda x: np.mean(x[:, 1])
Ds, Ns, avg_accept_est_mean = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.mean(x[:, 0])
_, _, avg_accept_mean = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 1], 25)
_, _, avg_accept_est_lower_25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 1], 75)
_, _, avg_accept_est_upper_25 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 1], 5)
_, _, avg_accept_est_lower_5 = gen_sparse_2d_array_from_dict(results, fun)
fun = lambda x: np.percentile(x[:, 1], 95)
_, _, avg_accept_est_upper_95 = gen_sparse_2d_array_from_dict(results, fun)
N_ind = np.where(Ns == N)[0][0]
plt.figure()
plt.plot(Ds, avg_accept_mean[:, N_ind], 'r')
plt.plot(Ds, avg_accept_est_mean[:, N_ind], 'b')
plt.plot(Ds, avg_accept_est_lower_25[:, N_ind], 'b-.')
plt.plot(Ds, avg_accept_est_lower_5[:, N_ind], color="grey")
plt.plot(Ds, avg_accept_est_upper_95[:, N_ind], color="grey")
plt.fill_between(Ds,
avg_accept_est_lower_5[:, N_ind],
avg_accept_est_upper_95[:, N_ind],
color="grey",
alpha=.5)
plt.plot(Ds, avg_accept_est_upper_25[:, N_ind], 'b-.')
plt.plot(Ds, avg_accept_mean[:, N_ind], 'r')
plt.xscale("log")
plt.grid(True)
plt.xlim([Ds.min(), Ds.max()])
plt.xlabel(r"$d$")
plt.title(r"n=%d" % N)
ylim = plt.ylim()
plt.ylim([ylim[0], 1.01])
plt.legend(["HMC", "KMC median", "KMC 25\%-75\%", "KMC 5\%-95\%"],
loc="lower left")
fname_base = fname.split(".")[-2]
plt.savefig(fname_base + "_N=%d.eps" % N, axis_inches='tight')
# plot_mcmc_result(result, D1=1, D2=6)
# compute how MMD evolves
for j, N in enumerate(Ns):
logger.info(
"MMD of %d benchmark samples against %d MCMC samples" %
(len(benchmark_samples), N))
samples_so_far = result.samples[warmup:(warmup + N)]
MMDs[i, j] = k.estimateMMD(benchmark_samples, samples_so_far)
med = np.median(MMDs, 0)
lower = np.percentile(MMDs, 20, 0)
upper = np.percentile(MMDs, 80, 0)
plt.plot(Ns, med, "-", color=colors[alg_idx])
plt.plot(Ns, lower, '--', color=colors[alg_idx])
plt.plot(Ns, upper, '--', color=colors[alg_idx])
# err = np.array([np.abs(med-lower),np.abs(med-upper)])
# plt.plot(Ns, med, color=colors[alg_idx])
# plt.errorbar(Ns, med, err, color=colors[alg_idx])
plt.yscale("log")
line1 = Line2D([0, 0], [0, 0], color=colors[0])
line2 = Line2D([0, 0], [0, 0], color=colors[1])
line3 = Line2D([0, 0], [0, 0], color=colors[2])
plt.legend((line1, line2, line3), ["KMC", "KAMH", "RW"])
plt.ylabel(r"MMD from ground truth")
plt.xlabel("Iterations")
plt.grid(True)