def test_ratio_plot_log(cmdopt, data_gen):
output = skh_plt.ratio_plot(dict(x=data_gen[0],
errorbars=True,
histtype='marker',
log=True,
err_x=False),
dict(x=data_gen[1],
weights=data_gen[2],
errorbars=True),
logx=True,
ratio_range=(0, 10))
if cmdopt == "generate":
with open(answer_dir + '/answers_ratio_plot_log.npz', 'wb') as f:
np.savez(f,
bc1=output[1][0],
be1=output[1][1],
bc2=output[2][0],
be2=output[2][1])
output[0][0].set_title('test_ratio_plot_log')
plt.show()
elif cmdopt == "test":
answers = np.load(answer_dir + '/answers_ratio_plot_log.npz')
assert (np.all(output[1][0] == answers['bc1']))
assert (np.all(output[1][1] == answers['be1']))
assert (np.all(output[2][0] == answers['bc2']))
assert (np.all(output[2][1] == answers['be2']))
def test_ratio_plot_stacked(cmdopt, data_gen):
output = skh_plt.ratio_plot(dict(x=[data_gen[0], data_gen[1]],
stacked=True,
errorbars=True),
dict(x=[data_gen[0], data_gen[1]],
weights=[data_gen[2], data_gen[2]],
stacked=True,
errorbars=True,
err_style='line'),
range=(-5, 5),
bins='blocks')
if cmdopt == "generate":
with open(answer_dir + '/answers_ratio_plot_stacked.npz', 'wb') as f:
np.savez(f,
bc1=output[1][0],
be1=output[1][1],
bc2=output[2][0],
be2=output[2][1])
output[0][0].set_title('test_ratio_plot_stacked')
plt.show()
elif cmdopt == "test":
answers = np.load(answer_dir + '/answers_ratio_plot_stacked.npz')
assert (np.all(output[1][0] == answers['bc1']))
assert (np.all(output[1][1] == answers['be1']))
assert (np.all(output[2][0] == answers['bc2']))
assert (np.all(output[2][1] == answers['be2']))
def test_ratio_plot_quick(cmdopt, data_gen):
# bin tests
with pytest.raises(KeyError):
skh_plt.ratio_plot(dict(x=data_gen[0], bins=10), dict(x=data_gen[1], bins=11))
output = skh_plt.ratio_plot(dict(x=data_gen[0]), dict(x=data_gen[1], bins=11))
assert(len(output[1][0]) == 11)
# range tests
with pytest.raises(KeyError):
skh_plt.ratio_plot(dict(x=data_gen[0], range=(0, 1)), dict(x=data_gen[1], range=(1, 2)))
output = skh_plt.ratio_plot(dict(x=data_gen[0], range=(-0.1, 0.1)), dict(x=data_gen[1]))
assert(output[1][1][0] >= -0.1 and output[1][1][-1] <= 0.1)
output = skh_plt.ratio_plot(dict(x=data_gen[0]), dict(x=data_gen[1], range=(-0.1, 0.1)))
assert(output[1][1][0] >= -0.1 and output[1][1][-1] <= 0.1)
def train_and_validate(steps=10000, minibatch=128, LRrange=[0.0001, 0.00001, 10000, 0], beta1=0.9, beta2=0.999, nafdim=16, depth=2, \
savedir='abcdnn', seed=100, retrain=False, train=True):
rawinputs, normedinputs, inputmeans, inputsigma, ncat_per_feature = prepdata(
)
print(ncat_per_feature)
inputdim = 4
ncat_per_feature = ncat_per_feature[0:inputdim]
conddim = normedinputs.shape[1] - inputdim
issignal = (rawinputs['njet'] >= 9) & (rawinputs['nbtag'] >= 3
) # signal_selection
isbackground = ~issignal
bkgnormed = normedinputs[isbackground]
bkg = rawinputs[isbackground]
xmax = np.reshape(inputmeans + 5 * inputsigma, inputmeans.shape[1])
m = ABCDdnn(ncat_per_feature, inputdim, minibatch=minibatch, conddim=conddim, LRrange=LRrange, \
beta1=beta1, beta2=beta2, nafdim=nafdim, depth=depth, savedir=savedir, retrain=retrain, seed=seed)
m.setrealdata(bkgnormed)
m.savehyperparameters()
m.monitorevery = 100
if train:
m.train(steps)
m.display_training()
nj9cut = True
if nj9cut:
ncol = 3 # for plots below
condlist = [[[
1.,
0.,
0.,
1.,
0.,
]], [[
0.,
1.,
0.,
1.,
0.,
]], [[
0.,
0.,
1.,
1.,
0.,
]], [[
1.,
0.,
0.,
0.,
1.,
]], [[
0.,
1.,
0.,
0.,
1.,
]], [[
0.,
0.,
1.,
0.,
1.,
]]]
select0 = (rawinputs['njet'] == 7) & (rawinputs['nbtag'] == 2)
select1 = (rawinputs['njet'] == 8) & (rawinputs['nbtag'] == 2)
select2 = (rawinputs['njet'] >= 9) & (rawinputs['nbtag'] == 2)
select3 = (rawinputs['njet'] == 7) & (rawinputs['nbtag'] >= 3)
select4 = (rawinputs['njet'] == 8) & (rawinputs['nbtag'] >= 3)
select5 = (rawinputs['njet'] >= 9) & (rawinputs['nbtag'] >= 3)
select_data = [select0, select1, select2, select3, select4, select5]
plottextlist = [
f'$N_j=7, N_b=2$', f'$N_j=8, N_b=2$', f'$N_j\geq 9, N_b=2$',
f'$N_j=7, N_b\geq 3$', f'$N_j=8, N_b\geq 3$',
f'$N_j\geq 9, N_b\geq 3$'
]
njlist = [7, 8, 9, 7, 8, 9]
nblist = [2, 2, 2, 3, 3, 3]
else:
ncol = 3 # for plots
condlist = [[[
0.,
1.,
0.,
0.,
1.,
0.,
]], [[
0.,
0.,
1.,
0.,
1.,
0.,
]], [[
0.,
0.,
0.,
1.,
1.,
0.,
]], [[
0.,
1.,
0.,
0.,
0.,
1.,
]], [[
0.,
0.,
1.,
0.,
0.,
1.,
]], [[
0.,
0.,
0.,
1.,
0.,
1.,
]]]
select0 = (rawinputs['njet'] == 8) & (rawinputs['nbtag'] == 2)
select1 = (rawinputs['njet'] == 9) & (rawinputs['nbtag'] == 2)
select2 = (rawinputs['njet'] >= 10) & (rawinputs['nbtag'] == 2)
select3 = (rawinputs['njet'] == 8) & (rawinputs['nbtag'] >= 3)
select4 = (rawinputs['njet'] == 9) & (rawinputs['nbtag'] >= 3)
select5 = (rawinputs['njet'] >= 10) & (rawinputs['nbtag'] >= 3)
select_data = [select0, select1, select2, select3, select4, select5]
plottextlist = [
f'$N_j=8, N_b=2$', f'$N_j=9, N_b=2$', f'$N_j\geq 10, N_b=2$',
f'$N_j=8, N_b\geq 3$', f'$N_j=9, N_b\geq 3$',
f'$N_j\geq 10, N_b\geq 3$'
]
njlist = [8, 9, 10, 8, 9, 10]
nblist = [2, 2, 2, 3, 3, 3]
# create fake data
fakedatalist = []
for cond, nj, nb in zip(condlist, njlist, nblist):
nmcbatches = int(bkgnormed.shape[0] / minibatch)
nmcremain = bkgnormed.shape[0] % minibatch
fakelist = []
cond_to_append = np.repeat(cond, minibatch, axis=0)
for _ib in range(nmcbatches):
xin = bkgnormed[_ib * minibatch:(_ib + 1) * minibatch, :inputdim]
xin = np.hstack(
(xin,
cond_to_append)) # append conditional to the feature inputs
xgen = m.model.predict(xin)
#xgen = m.generate_sample(cond)
fakelist.append(xgen)
# last batch
xin = bkgnormed[nmcbatches * minibatch:, :inputdim]
xin = np.hstack(
(xin,
np.repeat(cond, nmcremain,
axis=0))) # append conditional to the feature inputs
xgen = m.model.predict(xin)
fakelist.append(xgen)
# all data
fakedata = np.vstack(fakelist)
fakedata = fakedata * inputsigma[:, :inputdim] + inputmeans[:, :
inputdim]
nfakes = fakedata.shape[0]
fakedata = np.hstack((fakedata, np.array([nj]*nfakes).reshape((nfakes,1))\
, np.array([nb]*nfakes).reshape(nfakes,1) )
)
fakedatalist.append(fakedata)
labelsindices = [['MET', 'met', 0.0, xmax[0]], ['H_T', 'ht', 0.0, xmax[1]],\
['p_{T5}', 'pt5', 0.0, xmax[2]], ['p_{T6}', 'pt6', 0.0, xmax[3]]]
nbins = 20
runplots = True
if runplots:
yscales = ['log', 'linear']
for yscale in yscales:
for li in labelsindices:
pos = featurevars.index(li[1])
fig, ax = plt.subplots(2, ncol, figsize=(3 * ncol, 6))
iplot = 0
for fakedata, seld, plottext in zip(fakedatalist, select_data,
plottextlist):
input_data = rawinputs[seld]
# Make ratio plots
plotaxes = MplPlotter.ratio_plot(dict(x=input_data[li[1]], bins=nbins, range=(li[2], li[3]), errorbars=True, normed=True, histtype='marker'), \
dict(x=fakedata[:, pos], bins=nbins, range=(li[2], li[3]), errorbars=True, normed=True), ratio_range=(0.25, 1.9))
plotfig = plotaxes[0][0].get_figure()
plotaxes[0][0].set_yscale(yscale)
plotfig.set_size_inches(5, 5)
plotfig.savefig(
os.path.join(
savedir,
f'result_{li[1]}_{iplot}_{yscale}_ratio.pdf'))
# make matrix of plots
row = iplot // ncol
col = iplot % ncol
iplot += 1
plt.sca(ax[row, col])
ax[row, col].set_yscale(yscale)
ax[row, col].set_xlabel(f"${li[0]}$ (GeV)")
MplPlotter.hist(input_data[li[1]],
bins=nbins,
alpha=0.5,
range=(li[2], li[3]),
errorbars=True,
histtype='marker',
normed=True)
MplPlotter.hist(fakedata[:, pos],
bins=nbins,
alpha=0.5,
range=(li[2], li[3]),
errorbars=True,
normed=True)
MplPlotter.hist(bkg[li[1]],
bins=nbins,
alpha=0.5,
range=(li[2], li[3]),
histtype='step',
normed=True)
plt.text(0.6,
0.8,
plottext,
transform=ax[row, col].transAxes,
fontsize=10)
fig.tight_layout()
fig.savefig(
os.path.join(savedir,
f'result_matrix_{li[1]}_{yscale}.pdf'))
generatesigsample = True
if generatesigsample:
bkgsigfakedata = np.vstack(fakedatalist)
datadict = {}
for var, idx in zip(featurevars, range(len(featurevars))):
datadict[var] = bkgsigfakedata[:, idx]
writetorootfile(os.path.join(savedir, 'fakedata_NAF.root'), datadict)
pass
