def do_phifit_leastsq(pool):
from scipy.optimize import leastsq
new_nodes = pool.where("state == 'creating-a-partition'")
for n in new_nodes:
n.restraints = []
for c in pool.converter:
k0 = get_force_constant(n)
pos0 = n.internals.getcoord(c)
all_values = pool.coord_range(c)
if(isinstance(c, DihedralCoordinate)):
p0 = [pos0, 2, k0] # Initial guess for parameters
restraint_class = DihedralRestraint
elif(isinstance(c, LinearCoordinate)):
p0 = [pos0, pos0, 0.1, k0] # Initial guess for parameters
restraint_class = DistanceRestraint
else:
raise(Exception("Unkown Coordinate-Type"))
#phi_values = get_phi_contrib(all_values, n, active_nodes, c)
phi_values = get_phi_contrib(all_values, n, c)
phi_potential = get_phi_contrib_potential(all_values, n, c)
node_value = n.internals.getcoord(c)
node_index = np.argmin(np.square(c.sub(all_values, node_value)))
#phi_potential -= phi_potential[node_index] # gauge: set phi_potential[node] = 0
# contiguous function = smooth penalty-surface
def heaviside(x): return 1/(1 + np.exp(-500*x))
phi_on = heaviside(phi_values - 0.01)
def errfunc(p):
p[1:] = [ max(i, 0) for i in p[1:] ] #all but p[0] should be positiv
restraint = restraint_class.calc_energy(p, all_values)
#penalties = (phi_values+0.01)*(phi_potential - restraint) # weich
#penalties = (phi_values+0.001)*(phi_potential - restraint) # also weich
#penalties = (phi_values+0.1)*(phi_potential - restraint) # hard
diff = restraint - phi_potential
restr_too_high = heaviside(diff)
penalties = np.abs(diff)
penalties += 15*phi_on*restr_too_high*np.abs(diff)
penalties += 10*abs(restraint[node_index])
return(penalties)
p1, success = leastsq(errfunc, p0)
assert(success)
print("p1 = "+str(p1))
new_restraint = restraint_class(c.atoms, *p1)
n.restraints.append(new_restraint)
def do_phifit_leastsq(pool):
from scipy.optimize import leastsq
new_nodes = pool.where("state == 'creating-a-partition'")
k0 = pool.get_force_constant()
for n in new_nodes:
n.restraints = []
for c in pool.converter:
pos0 = n.internals.getcoord(c)
all_values = pool.coord_range(c)
if(isinstance(c, DihedralCoordinate)):
p0 = [pos0, 2, k0] # initial guess for parameters
restraint_class = DihedralRestraint
elif(isinstance(c, LinearCoordinate)):
p0 = [pos0, pos0, 0.1, k0] # initial guess for parameters
restraint_class = DistanceRestraint
else:
raise(Exception("Unkown Coordinate-Type"))
phi_values = get_phi_contrib(all_values, n, c)
phi_potential = get_phi_contrib_potential(all_values, n, c)
node_value = n.internals.getcoord(c)
node_index = np.argmin(np.square(c.sub(all_values, node_value)))
# contiguous function = smooth penalty-surface
def heaviside(x): return 1/(1 + np.exp(-500*x))
phi_on = heaviside(phi_values - 0.01)
def errfunc(p):
p[1:] = [ max(i, 0) for i in p[1:] ] # all but p[0] should be positive
restraint = restraint_class.calc_energy(p, all_values)
#penalties = (phi_values+0.01)*(phi_potential - restraint) # soft
#penalties = (phi_values+0.001)*(phi_potential - restraint) # soft
#penalties = (phi_values+0.1)*(phi_potential - restraint) # hard
diff = restraint - phi_potential
restr_too_high = heaviside(diff)
penalties = np.abs(diff)
penalties += 15*phi_on*restr_too_high*np.abs(diff)
penalties += 10*abs(restraint[node_index])
return(penalties)
p1, success = leastsq(errfunc, p0)
assert(success)
print("p1 = "+str(p1))
new_restraint = restraint_class(c.atoms, *p1)
n.restraints.append(new_restraint)
def do_phifit_switch(pool):
new_nodes = pool.where("state == 'creating-a-partition'")
for n in new_nodes:
n.restraints = []
for c in pool.converter:
# analyze phi for this coordinate
all_values = pool.coord_range(c, lin_slack=False)
all_phi_pot = get_phi_contrib_potential(all_values, n, c)
norm_phi_pot = abs(all_phi_pot - np.min(all_phi_pot)) # we normalize all_phi_pot to a minimum of zero
max_phi_pot = np.max(norm_phi_pot)
argmax_phi_pot = np.argmax(norm_phi_pot)
plateau = np.where( norm_phi_pot < (max_phi_pot*PLATEAU_THRESHOLD) )[0]
node_value = n.internals.getcoord(c)
if isinstance(c, DihedralCoordinate):
if len(plateau) == 0: # constant phi: no restraint
phi0 = dphi = k = 0.0
print "Constant phi (no restraint) for coordinate %s of node %s." % (c, n.name)
else:
dih_plateau = np.take( all_values, plateau )
dih_plateau_left = dih_plateau[0] # 1st edge of plateau
dih_plateau_right = dih_plateau[len(plateau)-1] # 2nd edge of plateau
#TODO idea from Ole: transfer plateau_left to -pi first, calculate everything, then transfer back... could get rid of the part below
if (dih_plateau_left == -np.pi) and (dih_plateau_right == np.pi): # if plateau transcends +180 degrees, adjust edges
for i in range(1, len(dih_plateau)):
if abs(abs(dih_plateau[i])-abs(dih_plateau[i-1])) >= PLATEAU_BREAK:
dih_plateau_right = dih_plateau[i-1]
dih_plateau_left = dih_plateau[i]
break
diff_node2right = abs(DihedralCoordinate.sub(dih_plateau_right, node_value))
diff_node2left = abs(DihedralCoordinate.sub(dih_plateau_left, node_value))
plateau_size = np.radians(len(dih_plateau))
if (diff_node2right > plateau_size) or (diff_node2left > plateau_size): # node_value not on plateau
#print "extension necessary"
if diff_node2right < diff_node2left:
old_dih_plateau_right = dih_plateau_right
#print "old dih_plateau_right: %f" % old_dih_plateau_right
dih_plateau_right = node_value # extend plateau to the right
#print "new dih_plateau_right: %f" % dih_plateau_right
plateau_size_diff = abs(DihedralCoordinate.sub(old_dih_plateau_right, dih_plateau_right))
else:
old_dih_plateau_left = dih_plateau_left
#print "old dih_plateau_left: %f" % old_dih_plateau_left
dih_plateau_left = node_value # extend plateau to the left
#print "new dih_plateau_left: %f" % dih_plateau_left
plateau_size_diff = abs(DihedralCoordinate.sub(old_dih_plateau_left, dih_plateau_left))
#print "old plateau size: %f" % np.degrees(plateau_size)
plateau_size += plateau_size_diff
#print "new plateau size: %f" % np.degrees(plateau_size)
#print "plateau_size_diff: %f" % np.degrees(plateau_size_diff)
dphi = plateau_size/2
phi0 = DihedralCoordinate.sub(dih_plateau_right, dphi)
# get k via Steigungsdreieck
dih_phi_pot_max = all_values[argmax_phi_pot]
flank1 = np.abs( dih_phi_pot_max - dih_plateau_left )
flank1 = np.minimum( flank1, 2*np.pi-flank1 )
flank2 = np.abs( dih_phi_pot_max - dih_plateau_right )
flank2 = np.minimum( flank2, 2*np.pi-flank2 )
k = max_phi_pot/max(flank1, flank2)
n.restraints.append( DihedralRestraint(c.atoms, phi0, dphi, k) )
elif isinstance(c, LinearCoordinate):
if len(plateau) == 0: # constant phi: no restraint
r0 = r1 = r2 = k = 0.0
print "Constant phi (no restraint) for coordinate %s of node %s." % (c, n.name)
else:
lin_plateau = np.take( all_values, plateau )
r0 = lin_plateau[0] # 1st edge of plateau
r1 = lin_plateau[len(plateau)-1] # 2nd edge of plateau
r2 = THE_BIG_NUMBER # = using only harmonic part of restraint
# node_value not on plateau
if node_value > r1:
r1 = node_value # extend plateau to the right
elif node_value < r0:
r0 = node_value # extend plateau to the left
# get k via Steigungsdreieck
lin_phi_pot_max = all_values[argmax_phi_pot]
flank1 = np.abs(lin_phi_pot_max - r0)
flank2 = np.abs(lin_phi_pot_max - r1)
k = max_phi_pot/max(flank1, flank2)
n.restraints.append( DistanceRestraint(c.atoms, r0, r1, r2, k) )
else:
raise(Exception("Unkown Coordinate-Type"))
def update(self, dummy=None):
if(not self.active):
return
self.spin_cluster.set_sensitive(False)
if(not self.board.selected_coord or not self.board.selected_node):
return
current_coord = self.board.selected_coord
#throwing away all axes (might me two because of twiny)
#But otherwise e.g. autoscaling behaves strange when twinx are added and removed
self.board.canvas.figure.clear()
axes1 = self.board.canvas.figure.gca() # create new axes
axes1.grid()
axes1.set_ylabel('Probability')
parent_name = "n/a"
if self.board.selected_node.parent:
parent_name = self.board.selected_node.parent.name
if(self.board.cb_show_title.get_active()):
axes1.set_title( "%s : %s (parent: %s)"%(self.board.selected_node.name, current_coord.label, parent_name) )
axes1.set_autoscalex_on(False)
axes1.set_ylim(0, 1)
axes1.set_autoscaley_on(False)
axes1.set_xlabel(current_coord.plot_label)
scale = current_coord.plot_scale
xvalues = self.board.pool.coord_range(current_coord)
(lower, upper) = (min(xvalues), max(xvalues))
axes1.set_xlim( scale(lower), scale(upper) )
axes2 = axes1.twinx()
axes2.set_zorder(axes1.get_zorder() + 1) #axes2 should receive the pick_events
#axes1.patch.set_visible(False) # hide the 'canvas' - makes axes2 visible
axes2.set_autoscalex_on(False)
axes2.set_autoscaley_on(True)
axes2.set_ylabel('Energy')
if(self.board.cb_show_colors.get_active()):
plotargs_nodemarker = { 'bbox' : {'facecolor':'red'} }
plotargs_childmarker = { 'bbox' : {'facecolor':'green'} }
plotargs_sampling = {'facecolor':'blue'}
plotargs_parent = {'facecolor':'red', 'alpha':0.1}
plotargs_restraint = {'linewidth':2, 'color':'red'}
plotargs_phi = {'linewidth':2, 'color':'magenta', 'linestyle':'--'}
plotargs_phipotential = {'linewidth':2, 'color':'dimgrey', 'linestyle':'-'}
plotargs_cluster = {'facecolor':'red', 'label':'cluster', 'alpha':0.8}
plotargs_refpoints = {'facecolor':'orange', 'label':'refpoints', 'alpha':0.7}
plotargs_direct = {'facecolor':'limegreen'}
plotargs_corrected = {'facecolor':'lightskyblue'}
else:
plotargs_nodemarker = { 'bbox' : {'facecolor':'white'} }
plotargs_childmarker = { 'bbox' : {'facecolor':'lightgrey'} }
plotargs_sampling = {'facecolor':'grey'}
plotargs_parent = {'facecolor':'grey', 'alpha':0.2}
plotargs_restraint = {'linewidth':4, 'color':'black'}
plotargs_phi = {'linewidth':4, 'color':'dimgrey', 'linestyle':'--'}
plotargs_phipotential = {'linewidth':4, 'color':'black', 'linestyle':'--'}
plotargs_cluster = {'facecolor':'lightgrey', 'label':'cluster', 'alpha':1.0}
plotargs_refpoints = {'facecolor':'black', 'label':'refpoints', 'alpha':0.8}
plotargs_direct = {'facecolor':'white'}
plotargs_corrected = {'facecolor':'black'}
# NodeMarker
children = self.board.selected_node.children
for n in reversed(self.board.pool):
if(n == self.board.pool.root):
continue
label_pos = scale( n.internals.getcoord(current_coord) )
text = str(self.board.pool.index(n))
if(self.cb_marker.get_active() and n==self.board.selected_node):
axes2.text(label_pos, 0.0, text, picker=5, **plotargs_nodemarker)
elif(self.cb_child_markers.get_active() and (n in children)):
axes2.text(label_pos, 0.0, text, picker=5, **plotargs_childmarker)
# Histogram Plot
for n in self.board.pool:
if(self.cb_sampling.get_active() and n==self.board.selected_node):
plotargs = {'label':'sampling'}
plotargs.update(plotargs_sampling)
elif(self.cb_parent_sampling.get_active() and n==self.board.selected_node.parent):
plotargs = {'label':'parent sampling'}
plotargs.update(plotargs_parent)
else:
continue
if(not n.has_trajectory): continue
#not using plt.hist() - it's doesn't allow scaling y-axis to 0..1
samples = scale(n.trajectory.getcoord(current_coord))
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues), num=50))
weights = None
if(self.cb_use_frameweights.get_active() and n.has_internals and n.has_restraints):
weights = n.frameweights
hist = np.histogram(samples, bins=edges, weights=weights)[0]
height = hist.astype('float') / np.max(hist)
width = np.diff(edges)
left = edges[:-1]
axes1.bar(left, height, width, **plotargs)
# Restraint-, Phi- and Phi-Potential Plot
#for n in self.board.pool.where("state != 'refined'"):
for n in self.board.pool.where("has_restraints"):
if(n != self.board.selected_node): continue
#node_value = n.internals.getcoord(current_coord)
if(self.cb_restraint.get_active()):
restraint = n.restraints[current_coord.index]
penalties = restraint.energy(xvalues)
axes2.plot(scale(xvalues), penalties, label="restraint", **plotargs_restraint)
if(self.cb_phi.get_active()):
yvalues = get_phi_contrib(xvalues, n, current_coord)
axes1.plot(scale(xvalues), yvalues, label='phi', **plotargs_phi)
if(self.cb_phi_potential.get_active()):
yvalues = get_phi_contrib_potential(xvalues, n, current_coord)
axes2.plot(scale(xvalues), yvalues, label="phi potential", **plotargs_phipotential)
# WeightedSamplingHistogram
if(self.cb_reweighted_hist.get_active()):
for rb in [self.rb_show_both, self.rb_show_direct, self.rb_show_corrected]:
rb.set_sensitive(True)
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues), num=50))
hist_direct = np.zeros(edges.size-1)
hist_corr = np.zeros(edges.size-1)
for n in self.board.pool.where("'weight_direct' in obs or 'weight_corrected' in obs"):
samples = scale( n.trajectory.getcoord(current_coord) )
hist_node = np.histogram(samples, bins=edges, weights=n.frameweights, normed=True)[0]
if('weight_direct' in n.obs):
hist_direct += n.obs.weight_direct * hist_node
if('weight_corrected' in n.obs):
hist_corr += n.obs.weight_corrected * hist_node
width = np.diff(edges)
left = edges[:-1]
if(np.max(hist_direct) > 0 and not self.rb_show_corrected.get_active()):
height_direct = hist_direct.astype('float') / np.max(hist_direct)
axes1.bar(left, height_direct, width/2, label="weighted direct", **plotargs_direct)
if(np.max(hist_corr) > 0 and not self.rb_show_direct.get_active()):
height_corr = hist_corr.astype('float') / np.max(hist_corr)
axes1.bar(left+width/2, height_corr, width/2, label="weighted corrected", **plotargs_corrected)
else:
for rb in [self.rb_show_both, self.rb_show_direct, self.rb_show_corrected]:
rb.set_sensitive(False)
# ClusterHistogram
if(self.cb_cluster.get_active()):
self.cb_intra_cluster.set_sensitive(True)
if(os.path.exists(self.board.pool.chi_mat_fn)):
chi_threshold = 1E-3
npz_file = np.load(self.board.pool.chi_mat_fn)
chi_mat = npz_file['matrix']
node_names = npz_file['node_names']
n_clusters = chi_mat.shape[1]
self.spin_cluster.set_sensitive(True)
self.spin_cluster.set_range(1, n_clusters)
i = int(self.spin_cluster.get_value())
# presort to make (intra-cluster) plot faster
relevant_nodes = node_names[np.argwhere(chi_mat[:,i-1] > chi_threshold)]
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues), num=50))
hist_cluster = np.zeros(edges.size-1)
hist_all = np.zeros(edges.size-1)
for (n, chi) in zip([n for n in self.board.pool if n.name in node_names], chi_mat[:,i-1]):
if n.name not in relevant_nodes:
if self.cb_intra_cluster.get_active():
continue
else:
samples = scale( n.trajectory.getcoord(current_coord) )
hist_node = np.histogram(samples, bins=edges, weights=n.frameweights, normed=True)[0]
hist_all += n.obs.weight_corrected * hist_node
else:
samples = scale( n.trajectory.getcoord(current_coord) )
hist_node = np.histogram(samples, bins=edges, weights=n.frameweights, normed=True)[0]
hist_all += n.obs.weight_corrected * hist_node
hist_cluster += n.obs.weight_corrected * hist_node * chi
if self.cb_intra_cluster.get_active():
hist_all = hist_cluster
width = np.diff(edges)
left = edges[:-1]
if(np.max(hist_cluster) > 0):
height_cluster = hist_cluster.astype('float') / np.max(hist_all)
axes1.bar(left, height_cluster, width, **plotargs_cluster)
max_val = scale(np.linspace(np.min(xvalues), np.max(xvalues), num=50))[np.argmax(hist_cluster)]
axes1.text(axes1.get_xlim()[0], -0.1, "max=%.4f"%max_val, ha='left', bbox=dict(boxstyle="round", fc="1.0"))
axes1.text(axes1.get_xlim()[0], 1.05, "#involved nodes=%d"%len(relevant_nodes), ha='left', bbox=dict(boxstyle="round", fc="1.0"))
weight = np.load(self.board.pool.qc_mat_fn)["weights"][i-1]
axes1.text(axes1.get_xlim()[1], -0.1, "weight=%.4f"%weight, ha='right', bbox=dict(boxstyle="round", fc="1.0"))
else:
self.cb_intra_cluster.set_sensitive(False)
# RefpointsHistogram
if(self.cb_refpoints.get_active()):
if('refpoints' in self.board.selected_node.obs):
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues), num=50))
refternals = self.board.selected_node.trajectory.getframes(self.board.selected_node.obs['refpoints'])
hist = np.histogram(scale(refternals.getcoord(current_coord)), bins=edges, weights=None)[0]
height = hist.astype('float') / np.max(hist)
width = np.diff(edges)
left = edges[:-1]
n_ref = len(self.board.selected_node.obs['refpoints'])
n_steps = len(self.board.selected_node.trajectory)
axes1.bar(left, height, width, **plotargs_refpoints)
axes1.text(axes1.get_xlim()[1], -0.1, "ratio=%d/%d=%.2f"%(n_ref, n_steps, float(n_ref)/float(n_steps)), ha='right', bbox=dict(boxstyle="round", fc="1.0"))
# update legend
if(self.board.cb_show_legend.get_active()):
handles = axes2.get_legend_handles_labels()[0] + axes1.get_legend_handles_labels()[0]
labels = axes2.get_legend_handles_labels()[1] + axes1.get_legend_handles_labels()[1]
if(len(handles) > 0):
self.board.canvas.figure.legend(handles, labels)
# redraw
self.board.canvas.draw_idle()
def update(self, dummy=None):
if (not self.active):
return
self.spin_cluster.set_sensitive(False)
if (not self.board.selected_coord or not self.board.selected_node):
return
current_coord = self.board.selected_coord
#throwing away all axes (might me two because of twiny)
#But otherwise e.g. autoscaling behaves strange when twinx are added and removed
self.board.canvas.figure.clear()
axes1 = self.board.canvas.figure.gca() # create new axes
axes1.grid()
axes1.set_ylabel('Probability')
parent_name = "n/a"
if self.board.selected_node.parent:
parent_name = self.board.selected_node.parent.name
if (self.board.cb_show_title.get_active()):
axes1.set_title("%s : %s (parent: %s)" %
(self.board.selected_node.name,
current_coord.label, parent_name))
axes1.set_autoscalex_on(False)
axes1.set_ylim(0, 1)
axes1.set_autoscaley_on(False)
axes1.set_xlabel(current_coord.plot_label)
scale = current_coord.plot_scale
xvalues = self.board.pool.coord_range(current_coord)
(lower, upper) = (min(xvalues), max(xvalues))
axes1.set_xlim(scale(lower), scale(upper))
axes2 = axes1.twinx()
axes2.set_zorder(axes1.get_zorder() +
1) #axes2 should receive the pick_events
#axes1.patch.set_visible(False) # hide the 'canvas' - makes axes2 visible
axes2.set_autoscalex_on(False)
axes2.set_autoscaley_on(True)
axes2.set_ylabel('Energy')
if (self.board.cb_show_colors.get_active()):
plotargs_nodemarker = {'bbox': {'facecolor': 'red'}}
plotargs_childmarker = {'bbox': {'facecolor': 'green'}}
plotargs_sampling = {'facecolor': 'blue'}
plotargs_parent = {'facecolor': 'red', 'alpha': 0.1}
plotargs_restraint = {'linewidth': 2, 'color': 'red'}
plotargs_phi = {
'linewidth': 2,
'color': 'magenta',
'linestyle': '--'
}
plotargs_phipotential = {
'linewidth': 2,
'color': 'dimgrey',
'linestyle': '-'
}
plotargs_cluster = {
'facecolor': 'red',
'label': 'cluster',
'alpha': 0.8
}
plotargs_refpoints = {
'facecolor': 'orange',
'label': 'refpoints',
'alpha': 0.7
}
plotargs_direct = {'facecolor': 'limegreen'}
plotargs_corrected = {'facecolor': 'lightskyblue'}
else:
plotargs_nodemarker = {'bbox': {'facecolor': 'white'}}
plotargs_childmarker = {'bbox': {'facecolor': 'lightgrey'}}
plotargs_sampling = {'facecolor': 'grey'}
plotargs_parent = {'facecolor': 'grey', 'alpha': 0.2}
plotargs_restraint = {'linewidth': 4, 'color': 'black'}
plotargs_phi = {
'linewidth': 4,
'color': 'dimgrey',
'linestyle': '--'
}
plotargs_phipotential = {
'linewidth': 4,
'color': 'black',
'linestyle': '--'
}
plotargs_cluster = {
'facecolor': 'lightgrey',
'label': 'cluster',
'alpha': 1.0
}
plotargs_refpoints = {
'facecolor': 'black',
'label': 'refpoints',
'alpha': 0.8
}
plotargs_direct = {'facecolor': 'white'}
plotargs_corrected = {'facecolor': 'black'}
# NodeMarker
children = self.board.selected_node.children
for n in reversed(self.board.pool):
if (n == self.board.pool.root):
continue
label_pos = scale(n.internals.getcoord(current_coord))
text = str(self.board.pool.index(n))
if (self.cb_marker.get_active() and n == self.board.selected_node):
axes2.text(label_pos,
0.0,
text,
picker=5,
**plotargs_nodemarker)
elif (self.cb_child_markers.get_active() and (n in children)):
axes2.text(label_pos,
0.0,
text,
picker=5,
**plotargs_childmarker)
# Histogram Plot
for n in self.board.pool:
if (self.cb_sampling.get_active()
and n == self.board.selected_node):
plotargs = {'label': 'sampling'}
plotargs.update(plotargs_sampling)
elif (self.cb_parent_sampling.get_active()
and n == self.board.selected_node.parent):
plotargs = {'label': 'parent sampling'}
plotargs.update(plotargs_parent)
else:
continue
if (not n.has_trajectory): continue
#not using plt.hist() - it's doesn't allow scaling y-axis to 0..1
samples = scale(n.trajectory.getcoord(current_coord))
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues),
num=50))
weights = None
if (self.cb_use_frameweights.get_active() and n.has_internals
and n.has_restraints):
weights = n.frameweights
hist = np.histogram(samples, bins=edges, weights=weights)[0]
height = hist.astype('float') / np.max(hist)
width = np.diff(edges)
left = edges[:-1]
axes1.bar(left, height, width, **plotargs)
# Restraint-, Phi- and Phi-Potential Plot
#for n in self.board.pool.where("state != 'refined'"):
for n in self.board.pool.where("has_restraints"):
if (n != self.board.selected_node): continue
#node_value = n.internals.getcoord(current_coord)
if (self.cb_restraint.get_active()):
restraint = n.restraints[current_coord.index]
penalties = restraint.energy(xvalues)
axes2.plot(scale(xvalues),
penalties,
label="restraint",
**plotargs_restraint)
if (self.cb_phi.get_active()):
yvalues = get_phi_contrib(xvalues, n, current_coord)
axes1.plot(scale(xvalues),
yvalues,
label='phi',
**plotargs_phi)
if (self.cb_phi_potential.get_active()):
yvalues = get_phi_contrib_potential(xvalues, n, current_coord)
axes2.plot(scale(xvalues),
yvalues,
label="phi potential",
**plotargs_phipotential)
# WeightedSamplingHistogram
if (self.cb_reweighted_hist.get_active()):
for rb in [
self.rb_show_both, self.rb_show_direct,
self.rb_show_corrected
]:
rb.set_sensitive(True)
edges = scale(np.linspace(np.min(xvalues), np.max(xvalues),
num=50))
hist_direct = np.zeros(edges.size - 1)
hist_corr = np.zeros(edges.size - 1)
for n in self.board.pool.where(
"'weight_direct' in obs or 'weight_corrected' in obs"):
samples = scale(n.trajectory.getcoord(current_coord))
hist_node = np.histogram(samples,
bins=edges,
weights=n.frameweights,
normed=True)[0]
if ('weight_direct' in n.obs):
hist_direct += n.obs.weight_direct * hist_node
if ('weight_corrected' in n.obs):
hist_corr += n.obs.weight_corrected * hist_node
width = np.diff(edges)
left = edges[:-1]
if (np.max(hist_direct) > 0
and not self.rb_show_corrected.get_active()):
height_direct = hist_direct.astype('float') / np.max(
hist_direct)
axes1.bar(left,
height_direct,
width / 2,
label="weighted direct",
**plotargs_direct)
if (np.max(hist_corr) > 0
and not self.rb_show_direct.get_active()):
height_corr = hist_corr.astype('float') / np.max(hist_corr)
axes1.bar(left + width / 2,
height_corr,
width / 2,
label="weighted corrected",
**plotargs_corrected)
else:
for rb in [
self.rb_show_both, self.rb_show_direct,
self.rb_show_corrected
]:
rb.set_sensitive(False)
# ClusterHistogram
if (self.cb_cluster.get_active()):
self.cb_intra_cluster.set_sensitive(True)
if (os.path.exists(self.board.pool.chi_mat_fn)):
chi_threshold = 1E-3
npz_file = np.load(self.board.pool.chi_mat_fn)
chi_mat = npz_file['matrix']
node_names = npz_file['node_names']
n_clusters = chi_mat.shape[1]
self.spin_cluster.set_sensitive(True)
self.spin_cluster.set_range(1, n_clusters)
i = int(self.spin_cluster.get_value())
# presort to make (intra-cluster) plot faster
relevant_nodes = node_names[np.argwhere(
chi_mat[:, i - 1] > chi_threshold)]
edges = scale(
np.linspace(np.min(xvalues), np.max(xvalues), num=50))
hist_cluster = np.zeros(edges.size - 1)
hist_all = np.zeros(edges.size - 1)
for (n, chi) in zip(
[n for n in self.board.pool if n.name in node_names],
chi_mat[:, i - 1]):
if n.name not in relevant_nodes:
if self.cb_intra_cluster.get_active():
continue
else:
samples = scale(
n.trajectory.getcoord(current_coord))
hist_node = np.histogram(samples,
bins=edges,
weights=n.frameweights,
normed=True)[0]
hist_all += n.obs.weight_corrected * hist_node
else:
samples = scale(n.trajectory.getcoord(current_coord))
hist_node = np.histogram(samples,
bins=edges,
weights=n.frameweights,
normed=True)[0]
hist_all += n.obs.weight_corrected * hist_node
hist_cluster += n.obs.weight_corrected * hist_node * chi
if self.cb_intra_cluster.get_active():
hist_all = hist_cluster
width = np.diff(edges)
left = edges[:-1]
if (np.max(hist_cluster) > 0):
height_cluster = hist_cluster.astype('float') / np.max(
hist_all)
axes1.bar(left, height_cluster, width, **plotargs_cluster)
max_val = scale(
np.linspace(np.min(xvalues), np.max(xvalues),
num=50))[np.argmax(hist_cluster)]
axes1.text(axes1.get_xlim()[0],
-0.1,
"max=%.4f" % max_val,
ha='left',
bbox=dict(boxstyle="round", fc="1.0"))
axes1.text(axes1.get_xlim()[0],
1.05,
"#involved nodes=%d" % len(relevant_nodes),
ha='left',
bbox=dict(boxstyle="round", fc="1.0"))
weight = np.load(self.board.pool.qc_mat_fn)["weights"][i - 1]
axes1.text(axes1.get_xlim()[1],
-0.1,
"weight=%.4f" % weight,
ha='right',
bbox=dict(boxstyle="round", fc="1.0"))
else:
self.cb_intra_cluster.set_sensitive(False)
# RefpointsHistogram
if (self.cb_refpoints.get_active()):
if ('refpoints' in self.board.selected_node.obs):
edges = scale(
np.linspace(np.min(xvalues), np.max(xvalues), num=50))
refternals = self.board.selected_node.trajectory.getframes(
self.board.selected_node.obs['refpoints'])
hist = np.histogram(scale(refternals.getcoord(current_coord)),
bins=edges,
weights=None)[0]
height = hist.astype('float') / np.max(hist)
width = np.diff(edges)
left = edges[:-1]
n_ref = len(self.board.selected_node.obs['refpoints'])
n_steps = len(self.board.selected_node.trajectory)
axes1.bar(left, height, width, **plotargs_refpoints)
axes1.text(axes1.get_xlim()[1],
-0.1,
"ratio=%d/%d=%.2f" %
(n_ref, n_steps, float(n_ref) / float(n_steps)),
ha='right',
bbox=dict(boxstyle="round", fc="1.0"))
# update legend
if (self.board.cb_show_legend.get_active()):
handles = axes2.get_legend_handles_labels(
)[0] + axes1.get_legend_handles_labels()[0]
labels = axes2.get_legend_handles_labels(
)[1] + axes1.get_legend_handles_labels()[1]
if (len(handles) > 0):
self.board.canvas.figure.legend(handles, labels)
# redraw
self.board.canvas.draw_idle()
