def prepare_data(branchinfo_file="vars.yaml", region="2j2b"):
with open(branchinfo_file, "r") as f:
branches = yaml.load(f, Loader=yaml.FullLoader)
branches = branches[region]
ttbar = from_pytables(f"/home/ddavis/ATLAS/data/h5s/ttbar_r{region}.h5",
label=0,
auxlabel=1)
tW_DR = from_pytables(f"/home/ddavis/ATLAS/data/h5s/tW_DR_r{region}.h5",
label=1,
auxlabel=1)
tW_DS = from_pytables(f"/home/ddavis/ATLAS/data/h5s/tW_DS_r{region}.h5",
label=1,
auxlabel=0)
tW_DR.keep_columns(branches)
ttbar.keep_columns(branches)
tW_DS.keep_columns(branches)
scale_weight_sum(tW_DS, ttbar)
scale_weight_sum(tW_DR, ttbar)
tW_DR.weights *= 50
tW_DS.weights *= 50
ttbar.weights *= 100
X = pd.concat([ttbar.df, tW_DR.df, tW_DS.df]).to_numpy()
w = np.concatenate([ttbar.weights, tW_DR.weights, tW_DS.weights])
y = np.concatenate(
[ttbar.label_asarray, tW_DR.label_asarray, tW_DS.label_asarray])
z = np.concatenate([
ttbar.auxlabel_asarray, tW_DR.auxlabel_asarray, tW_DS.auxlabel_asarray
])
return (X, y, w, z)
def get_combined():
ttbar = from_pytables("ttbar.h5", "ttbar", label=0)
tW_DR = from_pytables("tW_DR.h5", "tW_DR", label=0)
tW_DS = from_pytables("tW_DS.h5", "tW_DS", label=1)
scale_weight_sum(tW_DR, ttbar)
scale_weight_sum(tW_DS, ttbar)
tW_DR.weights *= 0.5
tW_DS.weights *= 0.5
print(ttbar.weights.sum(), tW_DR.weights.sum(), tW_DS.weights.sum())
return ttbar, tW_DR, tW_DS
def test_save_and_read():
ds.to_pytables("outfile.h5")
new_ds = from_pytables("outfile.h5", ds.name)
X1 = ds.df.to_numpy()
X2 = new_ds.df.to_numpy()
w1 = ds.weights
w2 = new_ds.weights
np.testing.assert_array_almost_equal(X1, X2, 6)
np.testing.assert_array_almost_equal(w1, w2, 6)
def test_auxweights():
branches = ["pT_lep1", "pT_lep2", "eta_lep1", "eta_lep2"]
ds1 = from_root(
["tests/data/test_file.root"],
name="myds",
branches=branches,
auxweights=["phi_lep1", "phi_lep2"],
)
ds2 = from_root(
["tests/data/test_file.root"],
name="ds2",
branches=branches,
auxweights=["phi_lep1", "phi_lep2"],
)
ds1.append(ds2)
dsa = from_root(
["tests/data/test_file.root"],
name="myds",
branches=branches,
auxweights=["phi_lep1", "phi_lep2"],
)
dsb = from_root(
["tests/data/test_file.root"],
name="ds2",
branches=branches,
auxweights=["phi_lep1", "phi_lep2"],
)
dsc = dsa + dsb
np.testing.assert_array_almost_equal(ds1.auxweights["phi_lep1"],
dsc.auxweights["phi_lep1"], 5)
dsc.change_weights("phi_lep2")
assert dsc.weight_name == "phi_lep2"
pl2 = uproot.open("tests/data/test_file.root")["WtLoop_nominal"].array(
"phi_lep2")
nw2 = uproot.open("tests/data/test_file.root")["WtLoop_nominal"].array(
"weight_nominal")
ds2.change_weights("phi_lep2")
np.testing.assert_array_almost_equal(ds2.weights, pl2, 5)
assert "phi_lep2" not in ds2.auxweights
assert "weight_nominal" in ds2.auxweights
ds2.to_pytables("outfile1.h5")
ds2pt = from_pytables("outfile1.h5", "ds2", weight_name="phi_lep2")
print(ds2pt.auxweights)
np.testing.assert_array_almost_equal(
ds2pt.auxweights["weight_nominal"].to_numpy(), nw2)
os.remove("outfile1.h5")
assert True
def prepare_data(region="3jHb", delete_datasets=True):
ttbar = from_pytables(
f"/Users/ddavis/Desktop/newfullkincomb/ttbar_r{region}.h5",
label=0,
auxlabel=1)
tW_DR = from_pytables(
f"/Users/ddavis/Desktop/newfullkincomb/tW_DR_r{region}.h5",
label=1,
auxlabel=1)
# tW_DS = from_pytables(f"/home/ddavis/ATLAS/data/h5s/tW_DS_r{region}.h5", label=1, auxlabel=0)
# tW_DR.keep_columns(branches)
# ttbar.keep_columns(branches)
# tW_DS.keep_columns(branches)
# scale_weight_sum(tW_DS, ttbar)
scale_weight_sum(tW_DR, ttbar)
tW_DR.weights *= 100
# tW_DS.weights *= 50
ttbar.weights *= 100
X = pd.concat([ttbar.df, tW_DR.df]).to_numpy() # , tW_DS.df]).to_numpy()
w = np.concatenate([ttbar.weights, tW_DR.weights]) # , tW_DS.weights])
y = np.concatenate([ttbar.label_asarray(),
tW_DR.label_asarray()]) # , tW_DS.label_asarray()])
z = np.concatenate([ttbar.auxlabel_asarray(),
tW_DR.auxlabel_asarray()
] # , tW_DS.auxlabel_asarray()]
)
if delete_datasets:
del ttbar
del tW_DR
print("returning data")
return (X, y, w, z)
def __init__(
self,
name='zero_jet',
base_directory='../h5',
signal_h5='sig_one_jet.h5',
signal_name='sig',
signal_tree='wt_DR_nominal',
signal_latex=r'H$\rightarrow\mu\mu$',
backgd_h5='bkg_zero_jet.h5',
backgd_name='bkg',
backgd_tree='tt_nominal',
backgd_latex=r'Data sideband',
weight_name='weight',
variables=['Z_PT_FSR', 'Z_Y_FSR', 'Muons_CosThetaStar', 'm_mumu'],
has_syst=False,
syssig_h5='tW_DS_2j2b.h5',
syssig_name='tW_DS_2j2b',
syssig_tree='tW_DS',
syssig_latex=r'$tW$ DS',
has_mass=True,
reg_variable='m_mumu',
reg_latex=r'm_\mu\mu',
):
self.name = name
self.signal_label, self.backgd_label, self.center_label, self.syssig_label = 1, 0, 1, 0
self.signal_latex, self.backgd_latex = signal_latex, backgd_latex
self.signal = from_pytables(signal_h5,
signal_name,
tree_name=signal_tree,
weight_name=weight_name,
label=self.signal_label,
auxlabel=self.center_label)
self.backgd = from_pytables(backgd_h5,
backgd_name,
tree_name=backgd_tree,
weight_name=weight_name,
label=self.backgd_label,
auxlabel=self.center_label)
self.signal.keep_columns(variables)
self.backgd.keep_columns(variables)
self.has_syst = has_syst
self.syssig_latex = None if not self.has_syst else syssig_latex
self.losses_test = {'L_gen': [], 'L_dis': [], 'L_diff': []}
self.losses_train = {'L_gen': [], 'L_dis': [], 'L_diff': []}
self.has_mass = has_mass
self.reg_variable = reg_variable
self.reg_latex = reg_latex
if self.has_syst:
self.syssig = from_pytables(syssig_h5,
syssig_name,
tree_name=syssig_tree,
weight_name=weight_name,
label=self.signal_label,
auxlabel=self.syssig_label)
self.syssig.keep_columns(variables)
# Append syssig to signal
self.signal.append(self.syssig)
# Equalise signal weights to background weights
scale_weight_sum(self.signal, self.backgd)
self.X_raw = np.concatenate(
[self.signal.df.to_numpy(),
self.backgd.df.to_numpy()])
scaler = StandardScaler()
self.X = scaler.fit_transform(self.X_raw)
self.y = np.concatenate(
[self.signal.label_asarray(),
self.backgd.label_asarray()])
self.z = np.concatenate(
[self.signal.auxlabel_asarray(),
self.backgd.auxlabel_asarray()])
if self.has_mass:
signal = from_pytables(signal_h5,
signal_name,
tree_name=signal_tree,
weight_name=weight_name,
label=self.signal_label,
auxlabel=self.center_label)
backgd = from_pytables(backgd_h5,
backgd_name,
tree_name=backgd_tree,
weight_name=weight_name,
label=self.backgd_label,
auxlabel=self.center_label)
signal.keep_columns([reg_variable])
backgd.keep_columns([reg_variable])
def normalise(df):
df[df > 200] = 200
return (df - 110) / 70.
self.z = np.concatenate([
normalise(signal.df).to_numpy(),
normalise(backgd.df).to_numpy()
])
self.w = np.concatenate([self.signal.weights, self.backgd.weights])
self.output_path = '/'.join([base_directory, self.describe()]) + '/'
if not os.path.exists(self.output_path):
os.makedirs(self.output_path)
print('\033[92m[INFO]\033[0m', self.describe(),
self.signal.df.__getitem__, self.backgd.df.__getitem__)
print('\033[92m[INFO]\033[0m', '-' * 20)
#store the content
with open(self.output_path + self.name + '_event.pkl', 'wb') as pkl:
pickle.dump(scaler, pkl)
#store the content
with open(self.output_path + self.name + '_event_py2.pkl',
'wb') as pkl:
pickle.dump(scaler, pkl, protocol=2)
import numpy as np
from sklearn.model_selection import KFold
from sklearn.metrics import roc_auc_score
import xgboost as xgb
from twaml.data import from_pytables
from twaml.data import scale_weight_sum
import matplotlib.pyplot as plt
ttbar = from_pytables("ttbar_1j1b.h5", "ttbar", label=0)
tW_DR = from_pytables("tW_DR_1j1b.h5", "tW_DR", label=1)
sow = ttbar.weights.sum() + tW_DR.weights.sum()
mwfl = sow * 0.01
scale_weight_sum(tW_DR, ttbar)
y = np.concatenate([tW_DR.label_asarray, ttbar.label_asarray])
X = np.concatenate([tW_DR.df.to_numpy(), ttbar.df.to_numpy()])
w = np.concatenate([tW_DR.weights, ttbar.weights])
folder = KFold(n_splits=3, shuffle=True, random_state=414)
ttbar_dist = []
tW_dist = []
tW_w_dist = []
ttbar_w_dist = []
roc_aucs = []
for train_idx, test_idx in folder.split(X):
X_train, X_test = X[train_idx], X[test_idx]
y_train, y_test = y[train_idx], y[test_idx]
w_train, w_test = w[train_idx], w[test_idx]
param = {"max_depth": 4, "n_estimators": 150, "min_child_weight": mwfl}