def main(n_particles=100, T=250):
ts, xs, ys = load_generated_traj()
apf = make_apf_obj()
res = apf.run(
n_particles,
ys,
)
xs_apf = res['xts']
aidx_apf = res['aidxs']
long_res = to_long_format(
xs_apf,
aidx_apf,
T=T,
)
fig, ax = visualize_genealogy(**long_res, window=5)
odir = osp.join(osp.dirname(osp.dirname(osp.abspath(__name__))), 'res')
if not osp.exists(odir):
mkdir(odir)
fig.savefig(osp.join(odir, 'ex2-gen-exe.png'))
def main(n_particles = 100, T = 250):
odir = osp.join(osp.dirname(osp.dirname(osp.abspath(__name__))),'res')
if not osp.exists(odir):
mkdir(odir)
ts, xs, ys = load_generated_traj()
apf = make_apf_obj()
res_wess = apf.run(n_particles,
ys,
resampler = 'systematic',
adaptive_resampling = True)
xs_apf_wess,_,aidxs_apf_wess,ness_wess = res_wess.values()
fig_ness,ax_ness = visualize_ness_comp(ness_wess,n_particles)
fig_ness.savefig(osp.join(odir,'ex2-ness-comp.png'))
n_uni_wess = get_unique_primary_ancestors(aidxs_apf_wess)
long_res = to_long_format(xs_apf_wess,
aidxs_apf_wess,
T = T,
)
fig_gen, ax_gen = visualize_genealogy(**long_res,
window = 5,
xs_true = xs,)
fig_gen.savefig(osp.join(odir,'ex2-gen-exg.png'))
def main(n_particles=100, T=250):
odir = osp.join(osp.dirname(osp.dirname(osp.abspath(__name__))), 'res')
if not osp.exists(odir):
mkdir(odir)
ts, xs, ys = load_generated_traj()
apf = make_apf_obj()
res_sys = apf.run(n_particles, ys, resampler='systematic')
xs_apf_sys = res_sys['xts']
aidxs_apf_sys = res_sys['aidxs']
res_mult = apf.run(n_particles, ys, resampler='multinomial')
xs_apf_mult = res_mult['xts']
aidxs_apf_mult = res_mult['aidxs']
n_uni_sys = get_unique_primary_ancestors(aidxs_apf_sys)
n_uni_mult = get_unique_primary_ancestors(aidxs_apf_mult)
fig_uni, ax_uni = visualize_diveristy(n_uni_sys,
n_uni_mult,
labels=['systematic', 'multinomial'])
fig_uni.savefig(osp.join(odir, 'ex2-sys-vs-mult.png'))
long_res = to_long_format(
xs_apf_sys,
aidxs_apf_sys,
T=T,
)
fig_gen, ax_gen = visualize_genealogy(**long_res, window=5)
fig_gen.savefig(osp.join(odir, 'ex2-gen-exf.png'))
def main():
ts, xs, ys = load_generated_traj()
KF = make_kf_obj()
xs_kf, ps_kf = KF.estimate_traj(ys, T=ts[-1] + 1)
_, Nsm, diff_bpf_mu, diff_bpf_var = load_bpf_results()
apf = make_apf_obj()
mu_results = {}
var_results = {}
table_results = np.zeros((Nsm.shape[0], 2))
for k, n_particles in enumerate(Nsm):
res = apf.run(n_particles, ys)
xs_apf = res['xts']
aidxs_apf = res['aidxs']
mu = np.sum(xs_apf * res['weights'], axis=1)
var = np.sum(res['weights'] * (xs_apf - mu.reshape(-1, 1))**2, axis=1)
#mu = xs_apf.mean(axis = 1)
#var = xs_apf.var(axis = 1)
mu_diff = np.abs(mu - xs_kf)
var_diff = np.abs(var - ps_kf)
table_results[k, 0] = np.mean(mu_diff)
table_results[k, 1] = np.mean(var_diff)
mu_results.update({'mu_' + str(n_particles): mu_diff})
var_results.update({'var_' + str(n_particles): var_diff})
odir = osp.join(osp.dirname(osp.dirname(osp.abspath(__name__))), 'res')
if not osp.exists(odir):
mkdir(odir)
comp_fig, comp_ax = visualize_difference_comp(
bpf_mu=diff_bpf_mu.mean(axis=0),
bpf_var=diff_bpf_var.mean(axis=0),
apf_mu=table_results[:, 0],
apf_var=table_results[:, 1],
ns=Nsm)
comp_fig.savefig(osp.join(odir, 'ex2-comp-bpf-apf.png'))
table_results = pd.DataFrame(table_results,
index=Nsm,
columns=['Mean', 'Variance'])
mu_results = pd.DataFrame(mu_results, index=ts)
var_results = pd.DataFrame(var_results, index=ts)
mu_results.to_csv(osp.join(odir, 'ex2-mu-difference-apf.tsv'),
sep='\t',
header=True,
index=True)
var_results.to_csv(osp.join(odir, 'ex2-var-difference-apf.tsv'),
sep='\t',
header=True,
index=True)
with open(osp.join(odir, 'ex2-stat-mean-table-apf.tex'), 'w+') as fopen:
ltx = table_results.to_latex(index=True,
bold_rows=True,
column_format='c|c|c')
ltx = ltx.replace('\\\\', '\\\\ \n \\hline')
fopen.writelines(ltx)
def main():
ts, xs, ys = load_generated_traj()
xt_xtm1 = lambda xtm1: st.norm(0.8 * xtm1, np.sqrt(0.5))
yt_xt = lambda xt: st.norm(2 * xt, np.sqrt(0.1))
x0 = st.norm(0, np.sqrt(2))
KF = make_kf_obj()
xs_kf, ps_kf = KF.estimate_traj(ys, T=ts[-1] + 1)
bspf = BSPF(
xt_xtm1_dist=xt_xtm1,
yt_xt_dist=yt_xt,
x0_dist=x0,
)
ns = np.array([10, 50, 100, 2000, 5000]).astype(int)
mu_results = {}
var_results = {}
table_results = np.zeros((ns.shape[0], 2))
width = ns.shape[0] * 3 + 0.4
height = 2 * 3 + 0.4
fig, ax = plt.subplots(2,
ns.shape[0],
sharey='row',
figsize=(width, height))
for aa in ax.flatten():
aa.yaxis.set_tick_params(labelleft=True)
for k, n in enumerate(ns):
res = bspf.run(ys, n)
mu = np.sum(res['xs'] * res['weights'], axis=0)
var = np.sum(res['weights'] * (res['xs'] - mu)**2, axis=0)
mu_diff = np.abs(mu - xs_kf)
var_diff = np.abs(var - ps_kf)
table_results[k, 0] = np.mean(mu_diff)
table_results[k, 1] = np.mean(var_diff)
mu_results.update({'mu_' + str(n): mu_diff})
var_results.update({'var_' + str(n): var_diff})
ax[0, k].plot(ts[1::], mu_diff[1::], color='black')
ax[1, k].plot(ts[1::], var_diff[1::], color='black')
ax[0, k].set_ylabel(r'$|\mu_{KF}-\mu_{PF}|$')
ax[1, k].set_ylabel(r'$|\mathrm{Var}_{KF}-\mathrm{Var}_{PF}|$')
ax[0, k].set_xlabel('Time')
ax[1, k].set_xlabel('Time')
ax[0, k].set_title(f"{n} Particles")
for pos in ['top', 'right']:
for r in range(ax.shape[0]):
ax[r, k].spines[pos].set_visible(False)
odir = osp.join(osp.dirname(osp.dirname(osp.abspath(__name__))), 'res')
if not osp.exists(odir):
mkdir(odir)
fig.savefig(osp.join(odir, 'ex2-kalman-vs-pf.png'))
mu_results = pd.DataFrame(mu_results, index=ts)
var_results = pd.DataFrame(var_results, index=ts)
table_results = pd.DataFrame(table_results,
index=ns,
columns=['Mean', 'Variance'])
mu_results.to_csv(osp.join(odir, 'ex2-mu-difference-bpf.tsv'),
sep='\t',
header=True,
index=True)
var_results.to_csv(osp.join(odir, 'ex2-var-difference-bpf.tsv'),
sep='\t',
header=True,
index=True)
with open(osp.join(odir, 'ex2-stat-mean-table.tex'), 'w+') as fopen:
ltx = table_results.to_latex(index=True,
bold_rows=True,
column_format='c|c|c')
ltx = ltx.replace('\\\\', '\\\\ \n \\hline')
fopen.writelines(ltx)