def make_plot(items,
figfile='',
xlabel='',
ylabel='',
x_log=False,
y_log=False,
labels=[],
title='',
ps=qu.PlotStyle('dark')):
plt.cla()
plt.clf()
fig, ax = plt.subplots(1, 1)
colors = ps.colors
#fig.patch.set_facecolor('white')
for i, item in enumerate(items):
label = None
if len(labels) >= i:
label = labels[i]
color = colors[i % len(colors)]
ax.plot(item, label=label, color=color)
if x_log: ax.set_xscale('log')
if y_log: ax.set_yscale('log')
ax.set_title(title)
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ax.legend()
if figfile != '':
plt.savefig(figfile, transparent=True, facecolor=ps.canv_plt)
plt.show()
def MetricPlot(model_history,
acc_range=(0.5,1.), loss_range=(0.,0.7),
acc_log=False, loss_log=False,
plotpath='/',
model_keys=[],
plotstyle=qu.PlotStyle('dark')):
if(model_keys == []): model_keys = list(model_history.keys())
for model_key in model_keys:
fig, ax = plt.subplots(1,2,figsize=(15,5))
keys = ['acc','val_acc']
lines = [model_history[model_key][key] for key in keys]
epochs = np.arange(len(lines[0])) + 1
pu.multiplot_common(
ax[0],
epochs,
lines,
keys,
y_min = acc_range[0],
y_max = acc_range[1],
y_log = acc_log,
xlabel = 'Epoch',
ylabel = 'Accuracy',
title='Model accuracy for {}'.format(model_key),
ps=plotstyle
)
keys = ['loss','val_loss']
lines = [model_history[model_key][key] for key in keys]
pu.multiplot_common(
ax[1],
epochs,
lines,
keys,
y_min = loss_range[0],
y_max = loss_range[1],
y_log = loss_log,
xlabel = 'Epoch',
ylabel = 'Loss',
title='Model loss for {}'.format(model_key),
ps=plotstyle
)
# add grids
for axis in ax.flatten():
axis.grid(True,color=plotstyle.grid_plt)
qu.SaveSubplots(fig, ax, ['accuracy_{}'.format(model_key), 'loss_{}'.format(model_key)], savedir=plotpath, ps=plotstyle)
plt.show()
return
def roc_plot(ax,
xlist,
ylist,
xlabel='False positive rate',
ylabel='True positive rate',
x_min=0,
x_max=1.1,
x_log=False,
y_min=0,
y_max=1.1,
y_log=False,
linestyles=[],
colorgrouping=-1,
extra_lines=[[[0, 1], [0, 1]]],
labels=[],
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[],
title='',
ps=qu.PlotStyle('dark'),
colors=None):
'''
Shortcut for making a ROC curve.
'''
multiplot(ax,
xlist,
ylist,
xlabel=xlabel,
ylabel=ylabel,
x_min=x_min,
x_max=x_max,
x_log=x_log,
y_min=y_min,
y_max=y_max,
y_log=y_log,
linestyles=linestyles,
colorgrouping=colorgrouping,
extra_lines=extra_lines,
labels=labels,
atlas_x=atlas_x,
atlas_y=atlas_y,
simulation=simulation,
textlist=textlist,
title=title,
ps=ps,
colors=colors)
return
def histogramOverlay(ax,
data,
labels,
xlabel,
ylabel,
x_min=0,
x_max=2200,
xbins=22,
normed=True,
y_log=False,
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[],
ps=qu.PlotStyle('dark')):
xbin = np.arange(x_min, x_max, (x_max - x_min) / xbins)
zorder_start = -1 * len(data) # hack to get axes on top
colors = ps.colors
for i, vals in enumerate(data):
ax.hist(vals,
bins=xbin,
density=normed,
alpha=0.5,
label=labels[i],
color=colors[i % len(colors)],
zorder=zorder_start + i)
ax.set_xlim(x_min, x_max)
if y_log: ax.set_yscale('log')
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ps.SetStylePlt(ax)
# TODO: find a way to replace this
#if atlas_x >= 0 and atlas_y >= 0:
#ampl.draw_atlas_label(atlas_x, atlas_y, simulation = simulation, fontsize = 18)
#drawLabels(fig, atlas_x, atlas_y, simulation, textlist) #TODO: fix for fig,ax implementation
ax.set_zorder = len(data) + 1 #hack to keep the tick marks up
legend = ax.legend(facecolor=ps.canv_plt)
for leg_text in legend.get_texts():
leg_text.set_color(ps.text_plt)
return
def MapStabilityTest(mapping_func,
b_vals=[0., .1, .5, 1., 1.0e14],
m_vals=[1.],
x=np.linspace(0.001, 4., 1000),
ps=qu.PlotStyle('dark'),
savedir='',
legend_size=-1):
mb_combos = list(itertools.product(b_vals, m_vals))
if (m_vals == [1.]):
forward_labels = ['f(x), b={:.1e}'.format(mb[0]) for mb in mb_combos]
reverse_labels = [
'g(f(x)), b={:.1e}'.format(mb[0]) for mb in mb_combos
]
else:
forward_labels = [
'f(x), b={:.1e}, m={:.1e}'.format(mb[0], mb[1]) for mb in mb_combos
]
reverse_labels = [
'g(f(x)), b={:.1e}, m={:.1e}'.format(mb[0], mb[1])
for mb in mb_combos
]
forward = [mapping_func(b, m).Forward(x) for (b, m) in mb_combos]
reverse = [
mapping_func(mb_combos[i][0], mb_combos[i][1]).Inverse(forward[i])
for i in range(len(forward))
]
#reverse = [mapping_func(b,m).Inverse(forward) for (b,m) in mb_combos]
fig, ax = plt.subplots(1, 2, figsize=(16, 6))
y_min, y_max = (np.min(x), np.max(x))
pu.multiplot_common(ax[0],
x,
forward,
forward_labels,
y_min=y_min,
y_max=y_max,
xlabel='x',
ylabel='y',
title='Forward Mapping',
ps=ps)
pu.multiplot_common(ax[1],
x,
reverse,
reverse_labels,
y_min=y_min,
y_max=y_max,
xlabel='x',
ylabel='y',
title='Reverse Mapping',
ps=ps)
if (legend_size > 0): plt.rc('legend', fontsize=legend_size)
plt.show()
savename = 'mapping_test.png'
if (savedir != ''):
savename = savedir + '/' + savename
fig.savefig(savename, transparent=True)
return
def multiplot_common(ax,
xcenter,
lines,
labels,
xlabel,
ylabel,
x_min=None,
x_max=None,
y_min=None,
y_max=None,
x_log=False,
y_log=False,
x_ticks=None,
linestyles=[],
colorgrouping=-1,
extra_lines=[],
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[],
title='',
ps=qu.PlotStyle('dark'),
colors=None):
'''
Creates a set of plots, on a common carrier "xcenter".
Draws the plots on a provided axis.
'''
if (x_min == None): x_min = np.min(xcenter)
if (x_max == None): x_max = np.max(xcenter)
if (y_min == None): y_min = np.minimum(0., np.min(np.column_stack(lines)))
if (y_max == None): y_max = 1.25 * np.max(np.column_stack(lines))
if (x_ticks != None): ax.xaxis.set_major_locator(plt.MaxNLocator(x_ticks))
for extra_line in extra_lines:
ax.plot(extra_line[0], extra_line[1], linestyle='--', color='black')
if (colors is None): colors = ps.colors
for i, line in enumerate(lines):
if len(linestyles) > 0:
linestyle = linestyles[i]
else:
linestyle = 'solid'
if colorgrouping > 0:
color = colors[int(np.floor(i / colorgrouping))]
else:
color = colors[i % len(colors)]
ax.plot(xcenter,
line,
label=labels[i],
linestyle=linestyle,
color=color)
if x_log: ax.set_xscale('log')
if y_log: ax.set_yscale('log')
ax.set_xlim(x_min, x_max)
ax.set_ylim(y_min, y_max)
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ax.set_title(title)
ps.SetStylePlt(ax)
#drawLabels(fig, atlas_x, atlas_y, simulation, textlist)
legend = ax.legend(facecolor=ps.canv_plt)
for leg_text in legend.get_texts():
leg_text.set_color(ps.text_plt)
return
def multiplot(ax,
xlist,
ylist,
xlabel='False positive rate',
ylabel='True positive rate',
x_min=0,
x_max=1.1,
x_log=False,
y_min=0,
y_max=1.1,
y_log=False,
linestyles=[],
colorgrouping=-1,
extra_lines=[],
labels=[],
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[],
title='',
ps=qu.PlotStyle('dark'),
colors=None):
'''
Creates a set of plots, from series of x and y values (does not use a common carrier).
Draws the plots on a provided axis.
'''
for extra_line in extra_lines:
ax.plot(extra_line[0],
extra_line[1],
linestyle='--',
color=ps.main_plt)
if (colors is None): colors = ps.colors
for i, (x, y) in enumerate(zip(xlist, ylist)):
if len(linestyles) > 0:
linestyle = linestyles[i]
else:
linestyle = 'solid'
if colorgrouping > 0:
color = colors[int(np.floor(i / colorgrouping))]
else:
color = colors[i % (len(colors))]
label = None
if len(labels) > 0:
label = labels[i]
ax.plot(x, y, label=label, linestyle=linestyle, color=color)
if x_log: ax.set_xscale('log')
if y_log: ax.set_yscale('log')
ax.set_title(title)
ax.set_xlim(x_min, x_max)
ax.set_ylim(y_min, y_max)
ax.set_xlabel(xlabel)
ax.set_ylabel(ylabel)
ps.SetStylePlt(ax)
legend = ax.legend(facecolor=ps.canv_plt)
for leg_text in legend.get_texts():
leg_text.set_color(ps.text_plt)
#drawLabels(fig, atlas_x, atlas_y, simulation, textlist)
return
def EnergySummary(train_dfs,
valid_dfs,
data_dfs,
energy_name,
model_name,
plotpath,
extensions=['png'],
plot_size=750,
full=True,
ps=qu.PlotStyle('dark'),
**kwargs):
ps.SetStyle()
max_energy = 2000. # GeV
max_energy_2d = max_energy
bin_energy = 300
ratio_range_2d = [0.3, 1.7]
bins_2d = [200, 70]
if ('max_energy' in kwargs.keys()):
max_energy = kwargs['max_energy']
max_energy_2d = max_energy
if ('max_energy_2d' in kwargs.keys()): max_energy = kwargs['max_energy_2d']
if ('bin_energy' in kwargs.keys()): bin_energy = kwargs['bin_energy']
if ('ratio_range_2d' in kwargs.keys()):
ratio_range_2d = kwargs['ratio_range_2d']
if ('bins_2d' in kwargs.keys()): bins_2d = kwargs['bins_2d']
# Dictionaries to keep track of all our histogram objects.
# Each entry will be a dictionary of hists.
# Outer key is data type (charged pion, neutral pion),
# inner key is data set (train, valid, all).
clusterE = {} # reco energy
clusterE_calib = {
} # cluster_ENG_CALIB_TOT (true energy, as far as we're concerned)
clusterE_pred = {} # predicted energy
#clusterE_true = {} # "truth" energy from the parton level (I think), not what we're after
clusterE_ratio1 = {} # ratio1: E_pred / ENG_CALIB_TOT
clusterE_ratio2 = {} # ratio2: E_reco / ENG_CALIB_TOT
clusterE_ratio2D = {} # ratio1 vs. ENG_CALIB_TOT
clusterE_ratio2D_zoomed = {} # ratio1 vs. ENG_CALIB_TOT (Zoomed on left)
ratio1_iqr = {} # IQR, from ratio1
ratio2_iqr = {} # IQR, from ratio2
ratio1_iqr_zoomed = {} # IQR, from ratio1
ratio2_iqr_zoomed = {} # IQR, from ratio2
# histogram stacks
energy_stacks = {}
iqr_stacks = {}
iqr_stacks_zoomed = {}
# keep track of mean/median curves from the 2D plots (one set for each).
mean_curves = {}
mean_curves_zoomed = {}
# keep track of our canvases, legends and histogram stacks
canvs = {}
legends = {}
key_conversions = {
'pp': '#pi^{#pm}',
'p0': '#pi^{0}',
}
dsets = {'train': train_dfs, 'valid': valid_dfs, 'all data': data_dfs}
for key in train_dfs.keys(
): # assuming all DataFrame dicts have the same keys
# Initialize the inner dictionaries.
clusterE[key] = {}
clusterE_calib[key] = {}
clusterE_pred[key] = {}
#clusterE_true[key] = {}
clusterE_ratio1[key] = {}
clusterE_ratio2[key] = {}
clusterE_ratio2D[key] = {}
clusterE_ratio2D_zoomed[key] = {}
ratio1_iqr[key] = {}
ratio2_iqr[key] = {}
ratio1_iqr_zoomed[key] = {}
ratio2_iqr_zoomed[key] = {}
energy_stacks[key] = {}
iqr_stacks[key] = {}
iqr_stacks_zoomed[key] = {}
mean_curves[key] = {}
mean_curves_zoomed[key] = {}
canvs[key] = {}
legends[key] = {}
iqr_stacks[key] = {}
for dkey, frame in dsets.items():
key2 = '(' + key_conversions[key] + ', ' + dkey + ')'
clusterE[key][dkey] = rt.TH1F(
qu.RN(), 'E_{reco} ' + key2 + '; E_{reco} [GeV];Count',
bin_energy, 0., max_energy)
clusterE_calib[key][dkey] = rt.TH1F(
qu.RN(),
'E_{calib}^{tot} ' + key2 + ';E_{calib}^{tot} [GeV];Count',
bin_energy, 0., max_energy)
clusterE_pred[key][dkey] = rt.TH1F(
qu.RN(), 'E_{pred} ' + key2 + ';E_{pred} [GeV];Count',
bin_energy, 0., max_energy)
#clusterE_true[key][dkey] = rt.TH1F(qu.RN(), 'E_{true} ' + key2 + ';E_{true} [GeV];Count', bin_energy,0.,max_energy)
clusterE_ratio1[key][dkey] = rt.TH1F(
qu.RN(), 'E_{pred} / E_{calib}^{tot} ' + key2 +
';E_{pred}/E_{calib}^{tot};Count', 250, 0., 10.)
clusterE_ratio2[key][dkey] = rt.TH1F(
qu.RN(), 'E / E_{calib}^{tot} ' + key2 +
';E_{reco}/E_{calib}^{tot]};Count', 250, 0., 10.)
qu.SetColor(clusterE[key][dkey], ps.main, alpha=0.4)
qu.SetColor(clusterE_calib[key][dkey], rt.kPink + 9, alpha=0.4)
qu.SetColor(clusterE_pred[key][dkey], ps.curve, alpha=0.4)
# qu.SetColor(clusterE_true[key][dkey], rt.kRed, alpha = 0.4)
qu.SetColor(clusterE_ratio1[key][dkey], ps.main, alpha=0.4)
qu.SetColor(clusterE_ratio2[key][dkey], ps.curve, alpha=0.4)
meas = frame[key]['clusterE'].to_numpy()
calib = frame[key]['cluster_ENG_CALIB_TOT'].to_numpy()
pred = frame[key][energy_name].to_numpy()
#true = frame[key]['truthE'].to_numpy()
ratio1 = pred / calib
ratio2 = meas / calib
for i in range(len(meas)):
clusterE[key][dkey].Fill(meas[i])
clusterE_calib[key][dkey].Fill(calib[i])
clusterE_pred[key][dkey].Fill(pred[i])
#clusterE_true[key][dkey].Fill(true[i])
clusterE_ratio1[key][dkey].Fill(ratio1[i])
clusterE_ratio2[key][dkey].Fill(ratio2[i])
# Fill the histogram stack for the energy ratios.
energy_stacks[key][dkey] = rt.THStack(
qu.RN(), clusterE_ratio1[key][dkey].GetTitle())
energy_stacks[key][dkey].Add(clusterE_ratio1[key][dkey])
energy_stacks[key][dkey].Add(clusterE_ratio2[key][dkey])
# Make the 2D energy ratio plots.
title = 'E_{pred}/E_{calib}^{tot} vs. E_{calib}^{tot} ' + key2 + ';E_{calib}^{tot} [GeV];E_{pred}/E_{calib}^{tot};Count'
x_range = [0., max_energy_2d]
nbins = bins_2d
mean_curves[key][dkey], clusterE_ratio2D[key][dkey] = EnergyPlot2D(
pred,
calib,
nbins=nbins,
x_range=x_range,
y_range=ratio_range_2d,
title=title,
offset=True)
title = 'E_{pred}/E_{calib}^{tot} vs. E_{calib}^{tot} ' + key2 + ';(E_{calib}^{tot} + 1) [GeV];E_{pred}/E_{calib}^{tot};Count'
x_range = [1., 1. + 0.01 * max_energy_2d]
nbins = bins_2d
nbins[0] = nbins[0] - 1
mean_curves_zoomed[key][dkey], clusterE_ratio2D_zoomed[key][
dkey] = EnergyPlot2D(pred,
calib,
nbins=nbins,
x_range=x_range,
y_range=ratio_range_2d,
title=title,
offset=False)
# Make the energy ratio IQR plots.
title = 'IQR(E_{x}/E_{calib}^{tot}) ' + key2 + ';E_{calib}^{tot} [GeV];IQR'
x_range = [0., max_energy_2d]
nbins = int(bins_2d[0] / 2)
ratio1_iqr[key][dkey] = IqrPlot(pred,
calib,
title=title,
nbins=nbins,
x_range=x_range)
ratio1_iqr[key][dkey].SetLineColor(ps.main)
ratio2_iqr[key][dkey] = IqrPlot(meas,
calib,
title=title,
nbins=nbins,
x_range=x_range)
ratio2_iqr[key][dkey].SetLineColor(ps.curve)
title = 'IQR(E_{x}/E_{calib}^{tot}) ' + key2 + ';(E_{calib}^{tot} + 1) [GeV];IQR'
x_range = [1., 1. + 0.01 * max_energy_2d]
nbins = int((bins_2d[0] - 1) / 2)
ratio1_iqr_zoomed[key][dkey] = IqrPlot(pred,
calib,
title=title,
nbins=nbins,
x_range=x_range,
offset=True)
ratio1_iqr_zoomed[key][dkey].SetLineColor(ps.main)
ratio2_iqr_zoomed[key][dkey] = IqrPlot(meas,
calib,
title=title,
nbins=nbins,
x_range=x_range,
offset=True)
ratio2_iqr_zoomed[key][dkey].SetLineColor(ps.curve)
# Fill the histogram stack for the IQR plots.
iqr_stacks[key][dkey] = rt.THStack(qu.RN(), title)
iqr_stacks[key][dkey].Add(ratio1_iqr[key][dkey])
iqr_stacks[key][dkey].Add(ratio2_iqr[key][dkey])
iqr_stacks_zoomed[key][dkey] = rt.THStack(qu.RN(), title)
iqr_stacks_zoomed[key][dkey].Add(ratio1_iqr_zoomed[key][dkey])
iqr_stacks_zoomed[key][dkey].Add(ratio2_iqr_zoomed[key][dkey])
# Prepare the list of plots we'll show (we might exclude some).
plots = [
clusterE, clusterE_calib, clusterE_pred, energy_stacks,
clusterE_ratio2D, clusterE_ratio2D_zoomed, iqr_stacks,
iqr_stacks_zoomed
]
if (not full):
plots = [
energy_stacks, clusterE_ratio2D, clusterE_ratio2D_zoomed,
iqr_stacks, iqr_stacks_zoomed
]
dkeys = list(dsets.keys())
# Make legend for the overlapping plots (1D energy ratios, and IQR plots)
legends[key] = rt.TLegend(0.7, 0.7, 0.85, 0.85)
legends[key].SetBorderSize(0)
legends[key].AddEntry(clusterE_ratio1[key][dkeys[0]], 'x = pred', 'f')
legends[key].AddEntry(clusterE_ratio2[key][dkeys[0]], 'x = reco', 'f')
nx = len(dkeys)
ny = len(plots)
canvs[key] = rt.TCanvas(qu.RN(), 'c_' + str(key), nx * plot_size,
ny * plot_size)
canvs[key].Divide(nx, ny)
for i, plot in enumerate(plots):
x = nx * i + 1
if (plot == energy_stacks or plot == iqr_stacks
or plot == iqr_stacks_zoomed):
for j, dkey in enumerate(dkeys):
canvs[key].cd(x + j)
draw_option = 'NOSTACK HIST'
if (plot != energy_stacks): draw_option = 'NOSTACK C'
plot[key][dkey].Draw(draw_option)
rt.gPad.SetGrid()
rt.gPad.SetLogy()
plot[key][dkey].GetHistogram().GetXaxis().SetTitle(
'E_{x}/E_{calib}^{tot}')
if (plot == energy_stacks):
plot[key][dkey].GetHistogram().GetYaxis().SetTitle(
clusterE_ratio1[key][dkey].GetYaxis().GetTitle())
# if(strat == 'jet'):
# plot[key][dkey].SetMinimum(5.0e-1)
# plot[key][dkey].SetMaximum(1.0e3)
plot[key][dkey].SetMinimum(5.0e-1)
plot[key][dkey].SetMaximum(2.0e5)
else:
plot[key][dkey].GetHistogram().GetYaxis().SetTitle(
ratio1_iqr[key][dkey].GetYaxis().GetTitle())
plot[key][dkey].SetMinimum(1.0e-2)
plot[key][dkey].SetMaximum(1.)
if (plot == iqr_stacks_zoomed):
rt.gPad.SetLogx()
rt.gPad.SetBottomMargin(0.15)
plot[key][dkey].GetXaxis().SetTitleOffset(1.5)
if (plot == iqr_stacks or plot == iqr_stacks_zoomed):
plot[key][dkey].SetMinimum(1.0e-3)
legends[key].SetTextColor(ps.text)
legends[key].Draw()
elif (plot == clusterE_ratio2D or plot == clusterE_ratio2D_zoomed):
for j, dkey in enumerate(dkeys):
canvs[key].cd(x + j)
plot[key][dkey].Draw('COLZ')
if (plot == clusterE_ratio2D):
mean_curves[key][dkey].Draw('SAME')
else:
mean_curves_zoomed[key][dkey].Draw('SAME')
rt.gPad.SetLogx()
rt.gPad.SetBottomMargin(0.15)
plot[key][dkey].GetXaxis().SetTitleOffset(1.5)
rt.gPad.SetLogz()
rt.gPad.SetRightMargin(0.2)
plot[key][dkey].GetXaxis().SetMaxDigits(4)
else:
for j, dkey in enumerate(dkeys):
canvs[key].cd(x + j)
plot[key][dkey].Draw('HIST')
plot[key][dkey].SetMinimum(5.0e-1)
rt.gPad.SetLogy()
# Draw the canvas
canvs[key].Draw()
# Save the canvas as a PDF & PNG image.
image_name = '_'.join([model_name, key, 'plots'])
for ext in extensions:
canvs[key].SaveAs(plotpath + image_name + '.' + ext)
results = {}
results['canv'] = canvs
results['plots'] = plots
results['curves'] = [mean_curves, mean_curves_zoomed]
results['legend'] = legends
return results
def RocCurves(model_scores,
data_labels,
roc_fpr, roc_tpr,
roc_thresh, roc_auc,
indices=[],
sample_weight=None,
plotpath = '/', plotname = 'ROC',
model_keys = [],
model_labels = [],
colors = [],
drawPlots=True, figsize=(15,5),
plotstyle=qu.PlotStyle('dark')):
if(model_keys == []): model_keys = list(model_scores.keys())
if(model_labels == []): model_labels = model_keys
if(colors == []): colors = plotstyle.colors
if(len(indices) != len(data_labels)): indices = np.full(len(data_labels), True, dtype=np.dtype('bool'))
for model_key in model_keys:
roc_fpr[model_key], roc_tpr[model_key], roc_thresh[model_key] = roc_curve(
data_labels[indices],
model_scores[model_key][indices],
drop_intermediate=False,
sample_weight=sample_weight
)
roc_auc[model_key] = auc(roc_fpr[model_key], roc_tpr[model_key])
print('Area under curve for {}: {}'.format(model_key, roc_auc[model_key]))
if(not drawPlots): return
# TODO: Sort model_keys by AUC
# Make a plot of the ROC curves
fig, ax = plt.subplots(1,2,figsize=figsize)
xlist = [roc_fpr[x] for x in model_keys]
ylist = [roc_tpr[x] for x in model_keys]
labels = []
for i,x in enumerate(model_keys):
labels.append('{} (area = {:.3f})'.format(model_labels[i], roc_auc[x]))
# labels = ['{} (area = {:.3f})'.format(x, roc_auc[x]) for x in model_keys]
title = 'ROC curve: classification of $\pi^+$ vs. $\pi^0$'
pu.roc_plot(ax[0],
xlist=xlist,
ylist=ylist,
labels=labels,
title=title,
ps=plotstyle,
colors=colors
)
#title = 'ROC curve (zoomed in at top left)'
pu.roc_plot(ax[1],
xlist=xlist,
ylist=ylist,
x_min=0. , x_max=0.25,
y_min=0.6, y_max=1.,
labels=labels,
title=title,
ps=plotstyle,
colors=colors
)
qu.SaveSubplots(fig, ax, [plotname, plotname + '_zoom'], savedir=plotpath, ps=plotstyle)
plt.show()
return
def ImagePlot(pcells, cluster, log=True, dynamic_range=False, layers=[], cell_shapes={}, scaled_shape = [], latex_mpl = {}, plotpath = '', filename = '', plotstyle=qu.PlotStyle('dark')):
# Set some default values.
if(layers == []): layers = list(mu.cell_meta.keys())
if(cell_shapes == {}): cell_shapes = {key: (val['len_eta'],val['len_phi']) for key,val in mu.cell_meta.items()}
if(latex_mpl == {}):
latex_mpl = {
'p0': '$\pi^{0}$',
'pp': '$\pi^{+}$'
}
scaling = False
if(scaled_shape != []): scaling = True
fig, ax = plt.subplots(len(pcells.keys()),len(layers),figsize=(60,20))
fig.patch.set_facecolor(plotstyle.canv_plt)
i = 0
for ptype, pcell in pcells.items():
for layer in layers:
axis = ax.flatten()[i]
# default behaviour: plot a single cluster
if(cluster >= 0): image = pcell[layer][cluster].reshape(cell_shapes[layer])
# if cluster index is negative, provide an average image
else: image = np.mean(pcell[layer],axis=0).reshape(cell_shapes[layer])
if(dynamic_range):
vmin, vmax = np.min(image), np.max(image)
vmax = np.maximum(np.abs(vmin),np.abs(vmax))
vmin = -vmax
if(vmax == 0. and vmin == 0.):
vmax = 0.1
vmin = -vmax
else: vmin, vmax = (-1.,1.)
norm = TwoSlopeNorm(vmin=vmin,vcenter=0.,vmax=vmax)
if(log): norm = SymLogNorm(linthresh = 0.001, linscale=0.001, vmin=vmin, vmax=vmax, base=10.)
cmap = plt.get_cmap('BrBG')
image = pcell[layer][cluster].reshape(cell_shapes[layer])
if(scaling):
# Use our ImageScaleBlock. It requires a 4d tensor, format is [batch,eta,phi,channel].
image = np.expand_dims(image, axis=(0,-1))
image = np.squeeze(ImageScaleBlock(new_shape=tuple(scaled_shape), normalization=True)([image]).numpy())
im = axis.imshow(
image,
extent=[-0.2, 0.2, -0.2, 0.2],
cmap=cmap,
origin='lower',
interpolation='nearest',
norm=norm
)
#axis.colorbar()
axis.set_title('{a} in {b}'.format(a=latex_mpl[ptype],b=layer))
axis.set_xlabel("$\Delta\phi$")
axis.set_ylabel("$\Delta\eta$")
plotstyle.SetStylePlt(axis)
divider = make_axes_locatable(axis)
cax = divider.append_axes('right', size='5%', pad=0.2)
cb = fig.colorbar(im, cax=cax, orientation='vertical')
plt.setp(plt.getp(cb.ax.axes, 'yticklabels'), color=plotstyle.text_plt)
i += 1
# show the plots
if(filename != ''): plt.savefig('{}/{}'.format(plotpath,filename),transparent=True,facecolor=plotstyle.canv_plt)
plt.show()
return
