def __call__(self, key, E, P):
k, ko, faus = key
ax = self.ax
if self.init_incomplete:
if self.plot_u or 'f' == faus:
self.init_incomplete = False
msft = abs(self.msft[key])
sprd = self.sprd[key]
if np.any(np.isinf(msft)):
not_available_text(ax, "Spectral stats not finite")
self.is_active = False
else:
self.line_msft, = ax.plot(msft, 'k', lw=2, label='Error')
self.line_sprd, = ax.plot(sprd,
'b',
lw=2,
label='Spread',
alpha=0.9)
ax.get_xaxis().set_major_locator(MaxNLocator(integer=True))
ax.legend()
else:
msft = abs(self.msft[key])
sprd = self.sprd[key]
self.line_sprd.set_ydata(sprd)
self.line_msft.set_ydata(msft)
# ax.set_ylim(*d_ylim(msft))
# ax.set_ylim(bottom=1e-5)
ax.set_ylim([1e-3, 1e1])
def __init__(self, fignum, stats, key0, plot_u, E, P, **kwargs):
fig, ax = place.freshfig(fignum, figsize=(6, 3))
ax.set_xlabel('Sing. value index')
ax.set_yscale('log')
self.init_incomplete = True
self.ax = ax
self.plot_u = plot_u
try:
self.msft = stats.umisf
self.sprd = stats.svals
except AttributeError:
self.is_active = False
not_available_text(ax, "Spectral stats not being computed")
def __call__(self, key, E, P):
k, ko, faus = key
if 'a' == faus:
w = self.stats.w[key]
N = len(w)
ax = self.ax
self.is_active = N < 10001
if not self.is_active:
not_available_text(ax, 'Not computed (N > threshold)')
return
counted = w > self.bins[0]
_ = [b.remove() for b in self.hist]
nn, _, self.hist = ax.hist(w[counted], bins=self.bins, color='b')
ax.set_ylim(top=max(nn))
ax.set_title(
'N: {:d}. N_eff: {:.4g}. Not shown: {:d}. '.format(
N, 1 / (w @ w), N - np.sum(counted)))
def __init__(self, fignum, stats, key0, plot_u, E, P, **kwargs):
GS = {'height_ratios': [4, 1], 'hspace': 0.09, 'top': 0.95}
fig, (ax, ax2) = place.freshfig(fignum,
figsize=(5, 6),
nrows=2,
gridspec_kw=GS)
if E is None and np.isnan(
P.diag if isinstance(P, CovMat) else P).all():
not_available_text(ax, ('Not available in replays'
'\ncoz full Ens/Cov not stored.'))
self.is_active = False
return
Nx = len(stats.mu[key0])
if Nx <= 1003:
C = np.eye(Nx)
# Mask half
mask = np.zeros_like(C, dtype=np.bool)
mask[np.tril_indices_from(mask)] = True
# Make colormap. Log-transform cmap,
# but not internally in matplotlib,
# so as to avoid transforming the colorbar too.
cmap = plt.get_cmap('RdBu_r')
trfm = mpl.colors.SymLogNorm(linthresh=0.2,
linscale=0.2,
base=np.e,
vmin=-1,
vmax=1)
cmap = cmap(trfm(np.linspace(-0.6, 0.6, cmap.N)))
cmap = mpl.colors.ListedColormap(cmap)
#
VM = 1.0 # abs(np.percentile(C,[1,99])).max()
im = ax.imshow(C, cmap=cmap, vmin=-VM, vmax=VM)
# Colorbar
_ = ax.figure.colorbar(im, ax=ax, shrink=0.8)
# Tune plot
plt.box(False)
ax.set_facecolor('w')
ax.grid(False)
ax.set_title("State correlation matrix:", y=1.07)
ax.xaxis.tick_top()
# ax2 = inset_axes(ax,width="30%",height="60%",loc=3)
line_AC, = ax2.plot(arange(Nx), ones(Nx), label='Correlation')
line_AA, = ax2.plot(arange(Nx), ones(Nx), label='Abs. corr.')
_ = ax2.hlines(0, 0, Nx - 1, 'k', 'dotted', lw=1)
# Align ax2 with ax
bb_AC = ax2.get_position()
bb_C = ax.get_position()
ax2.set_position([bb_C.x0, bb_AC.y0, bb_C.width, bb_AC.height])
# Tune plot
ax2.set_title("Auto-correlation:")
ax2.set_ylabel("Mean value")
ax2.set_xlabel("Distance (in state indices)")
ax2.set_xticklabels([])
ax2.set_yticks([0, 1] + list(ax2.get_yticks()[[0, -1]]))
ax2.set_ylim(top=1)
ax2.legend(frameon=True,
facecolor='w',
bbox_to_anchor=(1, 1),
loc='upper left',
borderaxespad=0.02)
self.ax = ax
self.ax2 = ax2
self.im = im
self.line_AC = line_AC
self.line_AA = line_AA
self.mask = mask
if hasattr(stats, 'w'):
self.w = stats.w
else:
not_available_text(ax)