def replot():
# res_size vs mc
ax1 = plt.subplot(211)
ax1.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
ax1.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
ax1.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of orthogonalization process on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=2)
plt.xlim([9,101])
# res_size vs lyap
ax2 = plt.subplot(212)
mcb_le_mean, mcb_le_std = get_mean_std(mcb_lap)
ax2.errorbar(reservoir_sizes, mcb_le_mean, yerr=mcb_le_std, label="before", fmt="--")
mca_le_mean, mca_le_std = get_mean_std(mca_lap)
ax2.errorbar(reservoir_sizes, mca_le_mean, yerr=mca_le_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("lyapunov exponent")
plt.grid(True)
ax2.legend(loc=2)
plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
fig = plt.figure()
ax = fig.add_subplot(111)
ax.errorbar(
range(memory_max),
np.average(forget_before, axis=0),
yerr=np.std(forget_before, axis=0),
label="before")
ax.errorbar(
range(memory_max),
np.average(forget_after, axis=0),
yerr=np.std(forget_after, axis=0),
label="after")
ax.text(45, 0.8, "MC = {:.2f}".format(mcbefore), color='blue')
ax.text(96, 0.8, "MC = {:.2f}".format(mcafter), color='green')
ax.grid(True)
ax.legend()
plt.title("MC forgetting curves before and after orthogonalization")
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext='pdf')
plt.show()
def replot():
together= True
if together:
max_yticks = 2
fig = plt.figure()
fig.subplots_adjust(hspace=0.4)
bins = np.linspace(0, 60, 60)
for i, sigma in enumerate(sigmas):
if together:
sp = plt.subplot(len(sigmas)*100 + 10 + 1 + i)
if i < len(sigmas) - 1:
plt.setp( sp.get_xticklabels(), visible=False)
yloc = plt.MaxNLocator(max_yticks)
sp.yaxis.set_major_locator(yloc)
#sp.yaxis.set_ticks(np.arange(70000,80000,2500))
plt.hist(hists[i], bins=bins, label="$\sigma={:0.3f}$".format(sigma), normed=True, alpha=0.5)
plt.legend(loc=2,prop={'size':16})
plt.grid(True)
plt.xlim([0, 60])
plt.yticks(fontsize=16)
#fig.set_size_inches(4, 3)
#plt.figure(figsize=(4,3))
fig = plt.gcf()
plt.xticks(fontsize=20)
#fig.set_size_inches(4, 3)
try_save_fig(ext="png")
try_save_fig(ext="eps")
#plt.savefig("figure3.eps")
plt.show()
def replot():
fig = plt.figure()
ax = fig.add_subplot(111)
ax.errorbar(range(memory_max),
np.average(forget_before, axis=0),
yerr=np.std(forget_before, axis=0),
label="before",
fmt=next(linecycler))
ax.errorbar(range(memory_max),
np.average(forget_after, axis=0),
yerr=np.std(forget_after, axis=0),
label="after",
fmt=next(linecycler))
ax.text(45, 0.8, "MC = {:.2f}".format(mcbefore), color='blue')
ax.text(96, 0.8, "MC = {:.2f}".format(mcafter), color='green')
ax.grid(True)
ax.legend()
plt.xlim((0, 140))
plt.ylim((-0.05, 1.2))
plt.title("forgetting curves for OG method")
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext='pdf')
plt.show()
def plot(Y, Yerr, savedir, xticks, lineticks, xlabel="", ylabel="", linelabels=[], loc=1, save=False):
xmax = np.zeros(len(lineticks))
ymax = np.zeros(len(lineticks))
for li, _ in enumerate(lineticks):
ymax[li] = np.max(Y[li])
xmax[li] = xticks[np.argmax(Y[li])]
if not linelabels:
linelabels = ["" for _ in lineticks]
for li, _ in enumerate(lineticks):
plt.errorbar(xticks, Y[li], yerr=Yerr[li], label=linelabels[li], fmt=next(linecycler))
#plt.text(xmax[li], ymax[li] + 3, "({:.3f}, {:.3f})".format(xmax[li], ymax[li]), fontsize=18, ha='center')
plt.xlabel(xlabel, fontsize=24, labelpad=-3)
plt.ylabel(ylabel, fontsize=24)
plt.xticks(fontsize=16)
plt.yticks(fontsize=16)
plt.grid(True)
plt.legend(loc=loc)
if save:
figname = os.path.join(savedir, 'figure')
try_save_fig(fname=figname)
try_save_fig(fname=figname, ext="pdf")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
ax1.errorbar(
xs,
np.average(MCs, axis=0),
yerr=np.std(MCs, axis=0))
plt.grid(True)
plt.ylabel("memory capacity")
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
plt.plot(xs, np.average(EVs, axis=0))
plt.grid(True)
plt.ylabel("abs(eigenvalues)")
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
plt.subplot(413, sharex=ax1)
plt.plot(xs, np.average(SVs, axis=0))
plt.grid(True)
plt.ylabel("singular value")
# lyapunov
plt.xlabel("iteration")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
# res_size vs mc
ax1 = plt.subplot(211)
ax1.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
ax1.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
ax1.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of orthonormalization process on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=2)
plt.xlim([9,101])
# res_size vs lyap
ax2 = plt.subplot(212)
ax2.errorbar(reservoir_sizes, mcb_l_mean, yerr=mcb_l_std, label="before", fmt="--")
ax2.errorbar(reservoir_sizes, mca_l_mean, yerr=mca_l_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("lyapunov exponent")
plt.grid(True)
ax2.legend(loc=2)
plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
fig = plt.figure()
ax = fig.add_subplot(111)
ax.errorbar(
range(memory_max),
np.average(forget_before, axis=0),
yerr=np.std(forget_before, axis=0),
label="before",
fmt=next(linecycler))
ax.errorbar(
range(memory_max),
np.average(forget_after, axis=0),
yerr=np.std(forget_after, axis=0),
label="after",
fmt=next(linecycler))
ax.text(45, 0.8, "MC = {:.2f}".format(mcbefore), color='blue')
ax.text(96, 0.8, "MC = {:.2f}".format(mcafter), color='green')
ax.grid(True)
ax.legend()
plt.xlim((0, 140))
plt.ylim((-0.05, 1.2))
plt.title("forgetting curves for OG method")
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext='pdf')
plt.show()
def compare_defs():
global q, sigma, tau, W, WI, mc
target_later = False
use_input = True
q = 20
sigma = 0.10
tau = 0.1
W = sp.random.normal(0, sigma, [q, q])
WI = sp.random.uniform(-tau, tau, q)
mc = memory_capacity(W,
WI,
memory_max=2 * q,
iterations_coef_measure=100000,
iterations=10000,
use_input=use_input,
target_later=target_later)
mcsum = sum(mc)
svds = svd(W, compute_uv=0)
print("MC = {}\nsmax={}\nsmin={}\nsspread={}".format(
mcsum, svds[0], svds[-1], svds[0] / svds[-1]))
plt.grid(True)
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
plt.title("starting k = {}; use_input = {}; MC = {:.3f}".format(
int(target_later), use_input, mcsum))
plt.plot(range(len(mc)), mc)
try_save_fig()
try_save_fig(ext="pdf")
plt.show()
def replot():
# res_size vs mc
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
for i in range(mcb_mean.shape[0]):
plt.errorbar(reservoir_sizes,
mcb_mean[i],
yerr=mcb_std[i],
label=rhos[i],
fmt="--",
color=cols[i])
for i in range(mcb_mean.shape[0]):
plt.errorbar(reservoir_sizes,
mca_mean[i],
yerr=mcb_std[i],
color=cols[i])
plt.ylabel("memory capacity")
plt.title("impact of ON method on MC")
plt.grid(True)
plt.legend(loc=4)
plt.xlabel("reservoir size")
# plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def scatter():
fig = plt.figure()
ax = fig.add_subplot(111)
xs = list(range(memory_max)) * INSTANCES
ax.scatter(
xs, forget_before.flatten(),
label="before", c='b')
ax.scatter(
xs, forget_after.flatten(),
label="after", c='g')
ax.text(5, 0.6, "MC = {:.2f}".format(mcbefore), color='blue')
ax.text(105, 0.6, "MC = {:.2f}".format(mcafter), color='green')
ax.grid(True)
ax.legend()
plt.title("MC forgetting curves before and after orthogonalization")
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
try_save_fig("figures/scatter")
try_save_fig("figures/scatter", ext='pdf')
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
ax1.errorbar(xs, np.average(MCs, axis=0), yerr=np.std(MCs, axis=0))
plt.grid(True)
plt.ylabel("memory capacity")
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
plt.plot(xs, np.average(EVs, axis=0))
plt.grid(True)
plt.ylabel("abs(eigenvalues)")
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
plt.subplot(413, sharex=ax1)
plt.plot(xs, np.average(SVs, axis=0))
plt.grid(True)
plt.ylabel("singular value")
# lyapunov
plt.xlabel("iteration")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
lims = [20, 50]
fs = 24
plt.subplot(121)
plt.xlabel("$\\sigma$", fontsize=fs)
plt.ylabel("MC")
plt.ylim(lims)
plt.grid(True)
plt.title("normal W, it={}".format(ITERATIONS))
plt.errorbar(sigmas, Y, Yerr)
plt.subplot(122)
plt.xlabel("$\\hat\\tau$", fontsize=fs)
plt.ylabel("MC")
plt.ylim(lims)
plt.grid(True)
plt.title("uniform W, it={}".format(ITERATIONS))
plt.errorbar(tauhats, Y2, Yerr2)
fig = plt.gcf()
fig.set_size_inches(14, 6)
try_save_fig(ext="png")
try_save_fig(ext="eps")
plt.show()
def compare_defs():
global q,sigma, tau, W, WI, mc
target_later = False
use_input = True
q = 20
sigma = 0.10
tau = 0.1
W = sp.random.normal(0, sigma, [q, q])
WI = sp.random.uniform(-tau, tau, q)
mc = memory_capacity(W, WI, memory_max=2*q,
iterations_coef_measure=100000, iterations=10000,
use_input=use_input, target_later=target_later)
mcsum = sum(mc)
svds = svd(W, compute_uv=0)
print("MC = {}\nsmax={}\nsmin={}\nsspread={}".format(mcsum, svds[0], svds[-1], svds[0] / svds[-1]))
plt.grid(True)
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
plt.title("starting k = {}; use_input = {}; MC = {:.3f}".format(
int(target_later), use_input, mcsum))
plt.plot(range(len(mc)), mc)
try_save_fig()
try_save_fig(ext="pdf")
plt.show()
def replot():
# res_size vs mc
ax1 = plt.subplot(211)
ax1.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
ax1.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
ax1.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of OG method on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=4)
plt.setp(ax1.get_xticklabels(), visible=False)
plt.xlim([9,101])
# res_size vs lyap
ax2 = plt.subplot(212)
ax2.errorbar(reservoir_sizes, mcb_l_mean, yerr=mcb_l_std, label="before", fmt="--")
ax2.errorbar(reservoir_sizes, mca_l_mean, yerr=mca_l_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("lyapunov exponent")
plt.grid(True)
ax2.legend(loc=4)
plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
# res_size vs mc
ax1 = plt.subplot(211)
ax1.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
ax1.errorbar(reservoir_sizes,
mcb_mean,
yerr=mcb_std,
label="before ",
fmt="--")
ax1.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of OG method on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=4)
plt.setp(ax1.get_xticklabels(), visible=False)
# plt.xlim([9,101])
# res_size vs lyap
ax2 = plt.subplot(212)
ax2.errorbar(reservoir_sizes,
mcb_l_mean,
yerr=mcb_l_std,
label="before",
fmt="--")
ax2.errorbar(reservoir_sizes, mca_l_mean, yerr=mca_l_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("Lyapunov exponent")
plt.grid(True)
ax2.legend(loc=4)
# plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
for li in range(len(sparsities)):
plt.errorbar(radiuses, Y[li], yerr=Yerr[li], label="sparsity = {:1.2f}".format(sparsities[li]))
plt.grid(True)
plt.legend(loc=3)
plt.xlabel("$|\lambda|_{max}$")
plt.ylabel("MC")
try_save_fig()
plt.show()
def gep_plot_lines(Y, Yerr, xticks, lineticks, xlabel="", ylabel="", loc=1, linelabel=""): for li in range(len(lineticks)): plt.errorbar(xticks, Y[li], yerr=Yerr[li], label=linelabel.format(lineval=lineticks[li])) plt.grid(True) plt.legend(loc=loc) plt.xlabel(xlabel) plt.ylabel(ylabel) try_save_fig() plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(311)
# for mc in MCs:
# plt.plot(xs, mc)
ax1.errorbar(xs,
np.average(MCs, axis=0),
yerr=np.std(MCs, axis=0))
plt.grid(True)
# plt.yticks([i for i in range(20, 102, 20)])
plt.ylabel("MC", size=20)
plt.ylim((30, 102))
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
e = []
ax2 = plt.subplot(312, sharex=ax1)
for ev in EVs:
plt.plot(xs, np.amax(ev, axis=1))
for ev in EVs:
e.append(np.amax(ev, axis=1))
# ax2.errorbar(xs,
# np.average(e, axis=0),
# yerr=np.std(e, axis=0))
# plt.yticks(np.arange(0.8, 1.01, 0.05))
plt.grid(True)
plt.ylabel(r'$\rho$', size=26)
plt.yticks(np.arange(0.76, 0.94, 0.04))
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(313, sharex=ax2)
s = []
# for sv in SVs:
# plt.plot(xs, np.amax(sv, axis=1))
for sv in SVs:
s.append(np.amax(sv, axis=1))
ax3.errorbar(xs,
np.average(s, axis=0),
yerr=np.std(s, axis=0))
plt.grid(True)
plt.ylabel(r'$s_{max}$', size=26)
plt.yticks(np.arange(1.0, 2.0, 0.2))
plt.setp(ax3.get_xticklabels(), visible=False)
plt.xlim((0, 60))
plt.xlabel("iteration", size=20)
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
for line in range(LINES):
plt.errorbar(sigmas, Y[line], yerr=Yerr[line], label=("res. size={0}".format(q)))
plt.grid(True)
plt.xlabel("sigma: $W = N(0, \\sigma)$")
plt.ylabel("MC, errbar: $1 \\times \\sigma$")
plt.legend(loc=3)
#plt.title("tau = {0}".format(tau))
try_save_fig()
plt.show()
def draw():
save = True
fig = plt.gcf()
fig.set_size_inches(8, 3)
plt.xticks(fontsize=18)
plt.yticks(fontsize=18)
if save:
try_save_fig()
try_save_fig(ext="pdf")
plt.clf()
else:
plt.show()
def draw(): save = True fig = plt.gcf() fig.set_size_inches(8,3) plt.xticks(fontsize=18) plt.yticks(fontsize=18) if save: try_save_fig() try_save_fig(ext="pdf") plt.clf() else: plt.show()
def replot():
for qi in range(LINES):
plt.errorbar(svals, Y[qi], label=("res. size={0}".format(res_sizes[qi])), yerr=Yerr[qi])
plt.plot(maxinlineX, maxinlineY, c=(0,0,0), label="maxima")
plt.grid(True)
plt.xlabel("$s_{max}$")
plt.ylabel("MC, errbar: $1 \\times \\sigma$")
plt.legend(loc=1)
#plt.title("tau = {0}".format(tau))
try_save_fig()
plt.show()
def replot():
for li in range(len(sparsities)):
plt.errorbar(radiuses,
Y[li],
yerr=Yerr[li],
label="sparsity = {:1.2f}".format(sparsities[li]))
plt.grid(True)
plt.legend(loc=3)
plt.xlabel("$|\lambda|_{max}$")
plt.ylabel("MC")
try_save_fig()
plt.show()
def replot():
for line in range(LINES):
plt.errorbar(sigmas,
Y[line],
yerr=Yerr[line],
label=("res. size={0}".format(q)))
plt.grid(True)
plt.xlabel("sigma: $W = N(0, \\sigma)$")
plt.ylabel("MC, errbar: $1 \\times \\sigma$")
plt.legend(loc=3)
#plt.title("tau = {0}".format(tau))
try_save_fig()
plt.show()
def r():
for li, dist in enumerate(dists):
interval = "input $\in [{}, {}]$".format(*dists[li][0])
plt.scatter(lines[li][0], lines[li][1],marker='+', c=dists[li][1], label=interval)
plt.legend(loc=3)
plt.grid(True)
plt.xlabel("LE")
plt.ylabel("MC")
plt.xlim([-1.1, 0.5])
plt.ylim([0, 30])
try_save_fig(ext="png")
try_save_fig(ext="eps")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(311)
# for mc in MCs:
# plt.plot(xs, mc)
ax1.errorbar(xs, np.average(MCs, axis=0), yerr=np.std(MCs, axis=0))
plt.grid(True)
# plt.yticks([i for i in range(20, 102, 20)])
plt.ylabel("MC", size=20)
plt.ylim((30, 102))
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
e = []
ax2 = plt.subplot(312, sharex=ax1)
for ev in EVs:
plt.plot(xs, np.amax(ev, axis=1))
for ev in EVs:
e.append(np.amax(ev, axis=1))
# ax2.errorbar(xs,
# np.average(e, axis=0),
# yerr=np.std(e, axis=0))
# plt.yticks(np.arange(0.8, 1.01, 0.05))
plt.grid(True)
plt.ylabel(r'$\rho$', size=26)
plt.yticks(np.arange(0.76, 0.94, 0.04))
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(313, sharex=ax2)
s = []
# for sv in SVs:
# plt.plot(xs, np.amax(sv, axis=1))
for sv in SVs:
s.append(np.amax(sv, axis=1))
ax3.errorbar(xs, np.average(s, axis=0), yerr=np.std(s, axis=0))
plt.grid(True)
plt.ylabel(r'$s_{max}$', size=26)
plt.yticks(np.arange(1.0, 2.0, 0.2))
plt.setp(ax3.get_xticklabels(), visible=False)
plt.xlim((0, 60))
plt.xlabel("iteration", size=20)
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
for qi in range(LINES):
plt.errorbar(svals,
Y[qi],
label=("res. size={0}".format(res_sizes[qi])),
yerr=Yerr[qi])
plt.plot(maxinlineX, maxinlineY, c=(0, 0, 0), label="maxima")
plt.grid(True)
plt.xlabel("$s_{max}$")
plt.ylabel("MC, errbar: $1 \\times \\sigma$")
plt.legend(loc=1)
#plt.title("tau = {0}".format(tau))
try_save_fig()
plt.show()
def replot():
# res_size vs mc
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
plt.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
plt.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of OG method on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=4)
plt.xlabel("reservoir size")
# plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
# res_size vs mc
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
plt.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
plt.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("memory capacity")
plt.title("impact of ON method on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=4)
plt.xlabel("reservoir size")
# plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
# res_size vs mc
sp1 = plt.subplot(211)
for i in range(mc.shape[0]):
sp1.plot(mc[i, :-1], label=reservoir_sizes[i])
plt.xlabel("time")
plt.ylabel("memory capacity")
plt.title("impact of OG method on MC")
plt.legend(loc=4)
sp2 = plt.subplot(212)
for i in range(mc.shape[0]):
sp2.plot(le[i, :-1], label=reservoir_sizes[i])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle="--")
plt.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ")
plt.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("memory capacity")
plt.title("impact of orthogonalization process on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=2)
print("tau={tau}, rho={rho}, eta={eta}".format(tau=tau, rho=rho, eta=eta))
print("instances={i}, ortho-iterations={oi}".format(i=INSTANCES, oi=ORTHOPROCESS_ITERATIONS))
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
# res_size vs mc
sp1 = plt.subplot(211)
for i in range(mc.shape[0]):
sp1.plot(mc[i, :-1], label=reservoir_sizes[i])
plt.xlabel("time")
plt.ylabel("memory capacity")
plt.title("impact of OG method on MC")
plt.legend(loc=4)
sp2 = plt.subplot(212)
for i in range(mc.shape[0]):
sp2.plot(le[i, :-1], label=reservoir_sizes[i])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
for si, sigma in enumerate(sigmas):
plt.plot(ks, mcks[si], label="$\\sigma = {:.3f}$".format(sigma))
fig = plt.gcf()
#plt.xticks(fontsize=20)
fig.set_size_inches(6, 4.5)
plt.grid(True)
plt.legend(loc=1)
plt.xlabel("$k$")
plt.ylabel("$MC_k$")
try_save_fig()
try_save_fig(ext="eps")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
ax1.errorbar(
xs,
np.average(MCs, axis=0),
yerr=np.std(MCs, axis=0))
plt.grid(True)
# plt.legend(loc=legend_location, labels='a')
plt.legend(loc=legend_location)
plt.ylabel("memory capacity")
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
plt.plot(xs, np.average(EVs, axis=0))
plt.grid(True)
plt.ylabel("abs(eigenvalues)")
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
plt.subplot(413, sharex=ax1)
plt.plot(xs, np.average(SVs, axis=0))
plt.grid(True)
plt.ylabel("singular value")
# lyapunov
ax4 = plt.subplot(414, sharex=ax1)
fixed_les = []
xs = list(range(0, 101, 10))
for i in range(len(LEs)):
fixed_les.append(LEs[i][xs])
print(fixed_les)
ax4.errorbar(
xs,
np.average(fixed_les, axis=0),
yerr=np.std(fixed_les, axis=0))
plt.grid(True)
plt.ylabel("lyapunov exponent")
plt.xlabel("iteration")
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
# res_size vs mc
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
for i in range(mcb_mean.shape[0]):
plt.errorbar(reservoir_sizes, mcb_mean[i], yerr=mcb_std[i], label=rhos[i], fmt="--", color=cols[i])
for i in range(mcb_mean.shape[0]):
plt.errorbar(reservoir_sizes, mca_mean[i], yerr=mcb_std[i], color=cols[i])
plt.ylabel("memory capacity")
plt.title("impact of ON method on MC")
plt.grid(True)
plt.legend(loc=4)
plt.xlabel("reservoir size")
# plt.xlim([9,101])
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
plt.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle="--")
plt.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ")
plt.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.xlabel("reservoir size")
plt.ylabel("memory capacity")
plt.title("impact of orthogonalization process on MC".format(tau, rho))
plt.grid(True)
plt.legend(loc=2)
print("tau={tau}, rho={rho}, eta={eta}".format(tau=tau, rho=rho, eta=eta))
print("instances={i}, ortho-iterations={oi}".format(
i=INSTANCES, oi=ORTHOPROCESS_ITERATIONS))
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
for mc in MCs:
plt.plot(xs, mc)
plt.grid(True)
# plt.yticks([i for i in range(20, 102, 20)])
plt.ylabel("MC", size=20)
plt.ylim((20, 102))
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
for ev in EVs:
plt.plot(xs, np.amax(ev, axis=1))
plt.grid(True)
# plt.ylabel("max(abs(eigenvalues))")
plt.ylabel(r'$\rho$', size=26)
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(413, sharex=ax1)
for sv in SVs:
plt.plot(xs, np.amax(sv, axis=1))
plt.grid(True)
plt.ylabel(r'$s_{max}$', size=26)
plt.setp(ax3.get_xticklabels(), visible=False)
# lyapunov
ax4 = plt.subplot(414, sharex=ax1)
for le in LEs:
plt.plot(xs, le)
plt.grid(True)
plt.ylabel('LE', size=26)
plt.xlim((0, 60))
plt.xlabel("iteration", size=20)
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
for mc in MCs:
plt.plot(xs, mc)
plt.grid(True)
# plt.yticks([i for i in range(20, 102, 20)])
plt.ylabel("MC", size=20)
plt.ylim((20, 102))
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
for ev in EVs:
plt.plot(xs, np.amax(ev, axis=1))
plt.grid(True)
# plt.ylabel("max(abs(eigenvalues))")
plt.ylabel(r'$\rho$', size=26)
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(413, sharex=ax1)
for sv in SVs:
plt.plot(xs, np.amax(sv, axis=1))
plt.grid(True)
plt.ylabel(r'$s_{max}$', size=26)
plt.setp(ax3.get_xticklabels(), visible=False)
# lyapunov
ax4 = plt.subplot(414, sharex=ax1)
for le in LEs:
plt.plot(xs, le)
plt.grid(True)
plt.ylabel('LE', size=26)
plt.xlim((0, 60))
plt.xlabel("iteration", size=20)
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def r():
for li, dist in enumerate(dists):
interval = "input $\in [{}, {}]$".format(*dists[li][0])
plt.scatter(lines[li][0],
lines[li][1],
marker='+',
c=dists[li][1],
label=interval)
plt.legend(loc=3)
plt.grid(True)
plt.xlabel("LE")
plt.ylabel("MC")
plt.xlim([-1.1, 0.5])
plt.ylim([0, 30])
try_save_fig(ext="png")
try_save_fig(ext="eps")
plt.show()
def draw(Y, Yerr, save=True):
global xmax, ymax
ymax = np.max(Y)
xmax = xticks[np.argmax(Y)]
plt.errorbar(xticks, Y, yerr=Yerr)
plt.xlabel(xlabel, fontsize=24, labelpad=-3)
plt.ylabel("memory capacity", fontsize=24)
#rc('xtick', labelsize=24)
#rc('ytick', labelsize=24)
plt.xticks(fontsize=16)
plt.yticks(fontsize=16)
plt.grid()
#plt.text(xmax, ymax, "({:.3f}, {:.3f})".format(xmax, ymax), fontsize=24)
if save:
try_save_fig()
try_save_fig(ext="pdf")
plt.show()
def draw(Y, Yerr, save=True):
global xmax, ymax
ymax = np.max(Y)
xmax = xticks[np.argmax(Y)]
plt.errorbar(xticks, Y, yerr=Yerr)
plt.xlabel(xlabel, fontsize=24, labelpad=-3)
plt.ylabel("memory capacity", fontsize=24)
#rc('xtick', labelsize=24)
#rc('ytick', labelsize=24)
plt.xticks(fontsize=16)
plt.yticks(fontsize=16)
plt.grid()
#plt.text(xmax, ymax, "({:.3f}, {:.3f})".format(xmax, ymax), fontsize=24)
if save:
try_save_fig()
try_save_fig(ext="pdf")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
ax1.errorbar(
xs,
np.average(MCs, axis=0),
yerr=np.std(MCs, axis=0))
plt.grid(True)
plt.ylabel("MC", size=20)
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
for ev in EVs:
plt.plot(xs, np.amax(ev, axis=1))
plt.grid(True)
# plt.ylabel("max(abs(eigenvalues))")
plt.ylabel(r'$\rho$', size=26)
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(413, sharex=ax1)
for sv in SVs:
plt.plot(xs, np.amax(sv, axis=1))
plt.grid(True)
plt.ylabel(r'$s_{max}$', size=26)
plt.setp(ax3.get_xticklabels(), visible=False)
# lyapunov
ax4 = plt.subplot(414, sharex=ax1)
ax4.errorbar(
xs,
np.average(LEs, axis=0),
yerr=np.std(LEs, axis=0))
plt.grid(True)
plt.ylabel("LE", size=20)
plt.xlabel("iteration", size=20)
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(411)
ax1.errorbar(xs, np.average(MCs, axis=0), yerr=np.std(MCs, axis=0))
plt.grid(True)
# plt.legend(loc=legend_location, labels='a')
plt.legend(loc=legend_location)
plt.ylabel("memory capacity")
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(412, sharex=ax1)
plt.plot(xs, np.average(EVs, axis=0))
plt.grid(True)
plt.ylabel("abs(eigenvalues)")
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
plt.subplot(413, sharex=ax1)
plt.plot(xs, np.average(SVs, axis=0))
plt.grid(True)
plt.ylabel("singular value")
# lyapunov
ax4 = plt.subplot(414, sharex=ax1)
fixed_les = []
xs = list(range(0, 101, 10))
for i in range(len(LEs)):
fixed_les.append(LEs[i][xs])
print(fixed_les)
ax4.errorbar(xs,
np.average(fixed_les, axis=0),
yerr=np.std(fixed_les, axis=0))
plt.grid(True)
plt.ylabel("lyapunov exponent")
plt.xlabel("iteration")
try_save_fig("figures/eta=" + str(eta))
try_save_fig("figures/eta=" + str(eta), ext="pdf")
plt.show()
def replot():
plt.figure(figsize=(8, 8))
ax = plt.subplot(1,1,1)
# res_size vs mc
for i in range(mc_mean.shape[1]):
ax.errorbar(reservoir_sizes, mc_mean[:, i], yerr=mc_std[:, i], label="{:.0e}".format(taus[i]))
ax.scatter(reservoir_sizes, mc_mean[:, i])
plt.ylabel("memory capacity")
# plt.xlim((99, 501))
plt.ylim((0, 500))
plt.title("ON method")
ax.grid(True)
handles, labels = ax.get_legend_handles_labels()
ax.legend(handles[::-1], labels[::-1], loc=2)
plt.xlabel("reservoir size")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def gep_plot_lines(Y,
Yerr,
xticks,
lineticks,
xlabel="",
ylabel="",
loc=1,
linelabel=""):
for li in range(len(lineticks)):
plt.errorbar(xticks,
Y[li],
yerr=Yerr[li],
label=linelabel.format(lineval=lineticks[li]))
plt.grid(True)
plt.legend(loc=loc)
plt.xlabel(xlabel)
plt.ylabel(ylabel)
try_save_fig()
plt.show()
def main():
global W, WI, U, L, V, L2
q = 100
sigma = 0.09
tau = 0.1
# boedecker's definition
target_later = True
use_input = False
W = sp.random.normal(0, sigma, [q, q])
WI = sp.random.uniform(-tau, tau, q)
U, L, V = svd(W)
smin = L[-1]
smax = L[0]
L2 = sp.zeros_like(L)
smins = sp.linspace(smin, 0.4, 40)
mcs = sp.zeros_like(smins)
for si, new_smin in enumerate(smins):
L2 = stretch_min(L, new_smin, smin, smax)
W2 = sp.dot(U, sp.dot(sp.diag(L2), V))
mc = memory_capacity(W2,
WI,
memory_max=2 * q,
iterations_coef_measure=1000,
iterations=1000,
use_input=use_input,
target_later=target_later)
mcs[si] = sum(mc)
plt.plot(smins, mcs)
plt.grid(True)
plt.xlabel("new $s_{min}$")
plt.ylabel("memory capacity")
try_save_fig()
try_save_fig(ext='pdf')
plt.show()
def plot(Y,
Yerr,
savedir,
xticks,
lineticks,
xlabel="",
ylabel="",
linelabels=[],
loc=1,
save=False):
xmax = np.zeros(len(lineticks))
ymax = np.zeros(len(lineticks))
for li, _ in enumerate(lineticks):
ymax[li] = np.max(Y[li])
xmax[li] = xticks[np.argmax(Y[li])]
if not linelabels:
linelabels = ["" for _ in lineticks]
for li, _ in enumerate(lineticks):
plt.errorbar(xticks,
Y[li],
yerr=Yerr[li],
label=linelabels[li],
fmt=next(linecycler))
#plt.text(xmax[li], ymax[li] + 3, "({:.3f}, {:.3f})".format(xmax[li], ymax[li]), fontsize=18, ha='center')
plt.xlabel(xlabel, fontsize=24, labelpad=-3)
plt.ylabel(ylabel, fontsize=24)
plt.xticks(fontsize=16)
plt.yticks(fontsize=16)
plt.grid(True)
plt.legend(loc=loc)
if save:
figname = os.path.join(savedir, 'figure')
try_save_fig(fname=figname)
try_save_fig(fname=figname, ext="pdf")
plt.show()
def replot():
# res_size vs mc
ax1 = plt.subplot(211)
ax1.plot(reservoir_sizes, reservoir_sizes, 'k', linestyle=":")
ax1.errorbar(reservoir_sizes, mcb_mean, yerr=mcb_std, label="before ", fmt="--")
ax1.errorbar(reservoir_sizes, mca_mean, yerr=mca_std, label="after")
plt.ylabel("MC")
plt.title("OG method")
plt.grid(True)
ax2 = plt.subplot(212)
ax2.errorbar(reservoir_sizes, leb_mean, yerr=leb_std, label="before ", fmt="--")
ax2.errorbar(reservoir_sizes, lea_mean, yerr=lea_std, label="after")
plt.ylabel("LE")
plt.grid(True)
plt.legend(loc=4)
plt.xlabel("reservoir size")
plt.yticks(np.arange(0.0, -0.11, -0.02))
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def replot():
together = True
if together:
max_yticks = 2
fig = plt.figure()
fig.subplots_adjust(hspace=0.4)
bins = np.linspace(0, 60, 60)
for i, sigma in enumerate(sigmas):
if together:
sp = plt.subplot(len(sigmas) * 100 + 10 + 1 + i)
if i < len(sigmas) - 1:
plt.setp(sp.get_xticklabels(), visible=False)
yloc = plt.MaxNLocator(max_yticks)
sp.yaxis.set_major_locator(yloc)
#sp.yaxis.set_ticks(np.arange(70000,80000,2500))
plt.hist(hists[i],
bins=bins,
label="$\sigma={:0.3f}$".format(sigma),
normed=True,
alpha=0.5)
plt.legend(loc=2, prop={'size': 16})
plt.grid(True)
plt.xlim([0, 60])
plt.yticks(fontsize=16)
#fig.set_size_inches(4, 3)
#plt.figure(figsize=(4,3))
fig = plt.gcf()
plt.xticks(fontsize=20)
#fig.set_size_inches(4, 3)
try_save_fig(ext="png")
try_save_fig(ext="eps")
#plt.savefig("figure3.eps")
plt.show()
def replot():
plt.figure(figsize=(8, 8))
ax = plt.subplot(1, 1, 1)
# res_size vs mc
for i in range(mc_mean.shape[1]):
ax.errorbar(reservoir_sizes,
mc_mean[:, i],
yerr=mc_std[:, i],
label="{:.0e}".format(taus[i]))
ax.scatter(reservoir_sizes, mc_mean[:, i])
plt.ylabel("memory capacity")
# plt.xlim((99, 501))
plt.ylim((0, 500))
plt.title("ON method")
ax.grid(True)
handles, labels = ax.get_legend_handles_labels()
ax.legend(handles[::-1], labels[::-1], loc=2)
plt.xlabel("reservoir size")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def main():
global W, WI, U, L, V, L2
q = 100
sigma = 0.09
tau = 0.1
# boedecker's definition
target_later = True
use_input = False
W = sp.random.normal(0, sigma, [q, q])
WI = sp.random.uniform(-tau, tau, q)
U, L, V = svd(W)
smin = L[-1]
smax = L[0]
L2 = sp.zeros_like(L)
smins = sp.linspace(smin, 0.4, 40)
mcs = sp.zeros_like(smins)
for si, new_smin in enumerate(smins):
L2 = stretch_min(L, new_smin, smin, smax)
W2 = sp.dot(U, sp.dot(sp.diag(L2), V))
mc = memory_capacity(W2, WI, memory_max=2*q,
iterations_coef_measure=1000, iterations=1000,
use_input=use_input, target_later=target_later)
mcs[si] = sum(mc)
plt.plot(smins, mcs)
plt.grid(True)
plt.xlabel("new $s_{min}$")
plt.ylabel("memory capacity")
try_save_fig()
try_save_fig(ext='pdf')
plt.show()
def replot():
xs = range(ORTHO_ITERATIONS + 1)
ax1 = plt.subplot(311)
plt.plot(xs, mc_gon, label="orthogonalization")
plt.plot(xs, mc_nor, label="orthonormalization")
plt.grid(True)
plt.legend(loc=legend_location)
#plt.xlabel("iteration")
plt.ylabel("memory capacity")
plt.setp(ax1.get_xticklabels(), visible=False)
# eigenvalues
ax2 = plt.subplot(312, sharex=ax1)
plt.plot(xs, eigenvaluesG, c='blue', label="gon")
plt.plot(xs, eigenvaluesN, c='green', label="norm")
plt.grid(True)
#plt.legend(loc=legend_location)
#plt.xlabel("iteration")
plt.ylabel("abs(eigenvalues)")
plt.setp(ax2.get_xticklabels(), visible=False)
# singular values
ax3 = plt.subplot(313, sharex=ax1)
plt.plot(xs, singular_valuesG, c='blue', label="gon")
plt.plot(xs, singular_valuesN, c='green', label="norm")
plt.grid(True)
#plt.legend(loc=legend_location)
plt.xlabel("iteration")
plt.ylabel("singular values")
try_save_fig("figures/figure")
try_save_fig("figures/figure", ext="pdf")
plt.show()
def r():
""" replot the values """
bar_width = 1
def barit(where, what, label, c):
av = np.average(what)
plt.bar([where], [av], bar_width, yerr=[np.std(what)], label=label, color=c)
plt.text(where + 0.5, av + 2, str(np.round(av, 2)), ha='center')
barit(bar_width*0, mcs_jaeger, "jaeger", 'r')
barit(bar_width*1, mcs_boedecker, "boedecker", 'b')
barit(bar_width*2, mcs_ja, "me", 'g')
barit(bar_width*3, mcs_other, "other", 'y')
# plt.bar([0], [np.average(mcs_jaeger)], bar_width, yerr=[np.std(mcs_jaeger)], label="jaeger")
# plt.bar([bar_width] , [np.average(mcs_boedecker)], bar_width, yerr=[np.std(mcs_boedecker)], label="boedecker")
# plt.bar([bar_width*2] , [np.average(mcs_ja)], bar_width, yerr=[np.std(mcs_ja)], label="ja")
# plt.bar([bar_width*3] , [np.average(mcs_other)], bar_width, yerr=[np.std(mcs_other)], label="other")
plt.legend(loc=3)
plt.grid(True)
plt.title("$\\tau={tau}$".format(tau=tau))
try_save_fig(ext="png")
try_save_fig(ext="eps")
plt.show()
