Mergiami

.gitignore

@@ -1,8 +1,10 @@
 *.pyc
 *.txt
 *.DS_Store
+*.zip
 
 plots/
+ana_overview_*/
 *.swp
 *.pdf
 *.png

NN/cfg/train_all.cfg.py → NN/cfg/train_all_3ycmbd.cfg.py

@@ -5,11 +5,9 @@
 from collections import OrderedDict
 from os import environ as env
 
-ch = 'mmm'
-set_paths(ch, 2018)
+set_paths('mmm', 2018) #FIXME this is just a dummy
 
 extra_selections = [
-    'hnl_2d_disp_sig>20',
     'hnl_pt_12>15',
     'sv_cos>0.99',
     'sv_prob>0.001',
@@ -87,6 +85,7 @@
                                       'hnl_m_12'   ,
                                       'sv_prob'    ,
                                       'hnl_2d_disp',
+                                      'year'       ,
                                       'channel'    ,],
 
                    composed_features = composed_features,
@@ -104,7 +103,6 @@
                    selection_mc_eem    = selection_eem + [cuts_eem.selections['is_prompt_lepton']],
 
                    selection_tight = cuts_mmm.selections_pd['tight'],
-                   lumi = 59700.,
 
                    epochs = 40,
                    )

NN/cfg/train_all_CR_MRloose.cfg.py

@@ -0,0 +1,120 @@
+import numpy as np
+from NN.nn_parametric_trainer import Trainer
+from plotter.selections import Selections
+from plotter.utils import set_paths
+from collections import OrderedDict
+from os import environ as env
+
+year = 2016
+
+set_paths('mmm', year) #FIXME channel is just a dummy
+
+extra_selections = [
+    'hnl_pt_12>15',
+    'sv_cos>0.90', ## MARTINA FOR STATS
+    # 'sv_prob>0.001', ## get rid of this for more stats, by MARTINA ...
+]
+
+cuts_mmm = Selections('mmm')
+selection_mmm = [ 
+    cuts_mmm.selections['pt_iso'], 
+    cuts_mmm.selections['baseline'], 
+    cuts_mmm.selections['vetoes_12_OS'], 
+    cuts_mmm.selections['vetoes_01_OS'], 
+    cuts_mmm.selections['vetoes_02_OS'],
+    cuts_mmm.selections['sideband'], 
+] + extra_selections
+
+cuts_mem = Selections('mem')
+selection_mem = [ 
+    cuts_mem.selections['pt_iso'], 
+    cuts_mem.selections['baseline'],
+    cuts_mem.selections['sideband'], 
+    cuts_mem.selections['vetoes_02_OS'],
+] + extra_selections
+
+cuts_eee = Selections('eee')
+selection_eee = [ 
+    cuts_eee.selections['pt_iso'], 
+    cuts_eee.selections['baseline'], 
+    cuts_eee.selections['vetoes_12_OS'], 
+    cuts_eee.selections['vetoes_01_OS'], 
+    cuts_eee.selections['vetoes_02_OS'],
+    cuts_eee.selections['sideband'], 
+] + extra_selections
+
+cuts_eem = Selections('eem')
+selection_eem = [ 
+    cuts_eem.selections['pt_iso'], 
+    cuts_eem.selections['baseline'],
+    cuts_eem.selections['sideband'], 
+    cuts_eem.selections['vetoes_01_OS'],
+] + extra_selections
+
+if year == 2017: 
+    selection_eee.append('l0_pt > 35')
+    selection_eem.append('l0_pt > 35')
+if year == 2018: 
+    selection_eee.append('l0_pt > 32')
+    selection_eem.append('l0_pt > 32')
+
+composed_features = OrderedDict()
+
+composed_features['abs_l0_eta'    ] = lambda df : np.abs(df.l0_eta)
+composed_features['abs_l1_eta'    ] = lambda df : np.abs(df.l1_eta)
+composed_features['abs_l2_eta'    ] = lambda df : np.abs(df.l2_eta)
+composed_features['log_abs_l0_dxy'] = lambda df : np.log10(np.abs(df.l0_dxy))
+composed_features['log_abs_l0_dz' ] = lambda df : np.log10(np.abs(df.l0_dz ))
+composed_features['log_abs_l1_dxy'] = lambda df : np.log10(np.abs(df.l1_dxy))
+composed_features['log_abs_l1_dz' ] = lambda df : np.log10(np.abs(df.l1_dz ))
+composed_features['log_abs_l2_dxy'] = lambda df : np.log10(np.abs(df.l2_dxy))
+composed_features['log_abs_l2_dz' ] = lambda df : np.log10(np.abs(df.l2_dz ))
+composed_features['abs_q_02'      ] = lambda df : np.abs(df.hnl_q_02)
+composed_features['abs_q_01'      ] = lambda df : np.abs(df.hnl_q_01)
+
+# https://stackoverflow.com/questions/20528328/numpy-logical-or-for-more-than-two-arguments
+# save a label to distinguish different channels
+# 1 = mmm
+# 2 = mem_os
+# 3 = mem_ss
+# 4 = eee
+# 5 = eem_os
+# 6 = eem_ss
+# composed_features['channel' ] = lambda df : 1 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==13 and np.abs(df.l2_pdgid)==13) + 2 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02!=0) + 3 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02==0) + 4 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==11) + 5 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02!=0) + 6 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02==0)
+
+trainer = Trainer (channel         = 'all_channels',
+                   base_dir        = env['NTUPLE_DIR'],
+                   #post_fix        = 'HNLTreeProducer_%s/tree.root' %ch,
+                   post_fix        = 'HNLTreeProducer/tree.root',
+
+                   years            = [year]       ,
+                   features        = ['l0_pt'      ,
+                                      'l1_pt'      ,
+                                      'l2_pt'      ,
+                                      'hnl_dr_12'  ,
+                                      'hnl_m_12'   ,
+                                      'sv_prob'    ,
+                                      'hnl_2d_disp',
+                                      'channel'    ,],
+
+                   composed_features = composed_features,
+
+                   selection_data_mmm  = selection_mmm,
+                   selection_mc_mmm    = selection_mmm + [cuts_mmm.selections['is_prompt_lepton']],
+
+                   selection_data_mem  = selection_mem,
+                   selection_mc_mem    = selection_mem + [cuts_mem.selections['is_prompt_lepton']],
+
+                   selection_data_eee  = selection_eee,
+                   selection_mc_eee    = selection_eee + [cuts_eee.selections['is_prompt_lepton']],
+
+                   selection_data_eem  = selection_eem,
+                   selection_mc_eem    = selection_eem + [cuts_eem.selections['is_prompt_lepton']],
+
+                   selection_tight = cuts_mmm.selections_pd['tight'],
+
+                   epochs = 20,
+                   )
+
+if __name__ == '__main__':
+    trainer.train()

NN/cfg/train_all_CR_bj.cfg.py

@@ -0,0 +1,120 @@
+import numpy as np
+from NN.nn_parametric_trainer import Trainer
+from plotter.selections import Selections
+from plotter.utils import set_paths
+from collections import OrderedDict
+from os import environ as env
+
+year = 2017
+
+set_paths('mmm', year) #FIXME channel is just a dummy
+
+extra_selections = [
+    'hnl_pt_12>15',
+    'sv_cos>0.90', ## MARTINA FOR STATS
+    # 'sv_prob>0.001', ## get rid of this for more stats, by MARTINA ...
+]
+
+cuts_mmm = Selections('mmm')
+selection_mmm = [ 
+    cuts_mmm.selections['pt_iso'], 
+    cuts_mmm.selections['baseline'], 
+    cuts_mmm.selections['vetoes_12_OS'], 
+    cuts_mmm.selections['vetoes_01_OS'], 
+    cuts_mmm.selections['vetoes_02_OS'],
+    cuts_mmm.selections['CR_bj'], 
+] + extra_selections
+
+cuts_mem = Selections('mem')
+selection_mem = [ 
+    cuts_mem.selections['pt_iso'], 
+    cuts_mem.selections['baseline'],
+    cuts_mem.selections['CR_bj'], 
+    cuts_mem.selections['vetoes_02_OS'],
+] + extra_selections
+
+cuts_eee = Selections('eee')
+selection_eee = [ 
+    cuts_eee.selections['pt_iso'], 
+    cuts_eee.selections['baseline'], 
+    cuts_eee.selections['vetoes_12_OS'], 
+    cuts_eee.selections['vetoes_01_OS'], 
+    cuts_eee.selections['vetoes_02_OS'],
+    cuts_eee.selections['CR_bj'], 
+] + extra_selections
+
+cuts_eem = Selections('eem')
+selection_eem = [ 
+    cuts_eem.selections['pt_iso'], 
+    cuts_eem.selections['baseline'],
+    cuts_eem.selections['CR_bj'], 
+    cuts_eem.selections['vetoes_01_OS'],
+] + extra_selections
+
+if year == 2017: 
+    selection_eee.append('l0_pt > 35')
+    selection_eem.append('l0_pt > 35')
+if year == 2018: 
+    selection_eee.append('l0_pt > 32')
+    selection_eem.append('l0_pt > 32')
+
+composed_features = OrderedDict()
+
+composed_features['abs_l0_eta'    ] = lambda df : np.abs(df.l0_eta)
+composed_features['abs_l1_eta'    ] = lambda df : np.abs(df.l1_eta)
+composed_features['abs_l2_eta'    ] = lambda df : np.abs(df.l2_eta)
+composed_features['log_abs_l0_dxy'] = lambda df : np.log10(np.abs(df.l0_dxy))
+composed_features['log_abs_l0_dz' ] = lambda df : np.log10(np.abs(df.l0_dz ))
+composed_features['log_abs_l1_dxy'] = lambda df : np.log10(np.abs(df.l1_dxy))
+composed_features['log_abs_l1_dz' ] = lambda df : np.log10(np.abs(df.l1_dz ))
+composed_features['log_abs_l2_dxy'] = lambda df : np.log10(np.abs(df.l2_dxy))
+composed_features['log_abs_l2_dz' ] = lambda df : np.log10(np.abs(df.l2_dz ))
+composed_features['abs_q_02'      ] = lambda df : np.abs(df.hnl_q_02)
+composed_features['abs_q_01'      ] = lambda df : np.abs(df.hnl_q_01)
+
+# https://stackoverflow.com/questions/20528328/numpy-logical-or-for-more-than-two-arguments
+# save a label to distinguish different channels
+# 1 = mmm
+# 2 = mem_os
+# 3 = mem_ss
+# 4 = eee
+# 5 = eem_os
+# 6 = eem_ss
+# composed_features['channel' ] = lambda df : 1 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==13 and np.abs(df.l2_pdgid)==13) + 2 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02!=0) + 3 * (np.abs(df.l0_pdgid)==13 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02==0) + 4 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==11) + 5 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02!=0) + 6 * (np.abs(df.l0_pdgid)==11 and np.abs(df.l1_pdgid)==11 and np.abs(df.l2_pdgid)==13 and df.hnl_q_02==0)
+
+trainer = Trainer (channel         = 'all_channels',
+                   base_dir        = env['NTUPLE_DIR'],
+                   #post_fix        = 'HNLTreeProducer_%s/tree.root' %ch,
+                   post_fix        = 'HNLTreeProducer/tree.root',
+
+                   years           = [year]        ,
+                   features        = ['l0_pt'      ,
+                                      'l1_pt'      ,
+                                      'l2_pt'      ,
+                                      'hnl_dr_12'  ,
+                                      'hnl_m_12'   ,
+                                      'sv_prob'    ,
+                                      'hnl_2d_disp',
+                                      'channel'    ,],
+
+                   composed_features = composed_features,
+
+                   selection_data_mmm  = selection_mmm,
+                   selection_mc_mmm    = selection_mmm + [cuts_mmm.selections['is_prompt_lepton']],
+
+                   selection_data_mem  = selection_mem,
+                   selection_mc_mem    = selection_mem + [cuts_mem.selections['is_prompt_lepton']],
+
+                   selection_data_eee  = selection_eee,
+                   selection_mc_eee    = selection_eee + [cuts_eee.selections['is_prompt_lepton']],
+
+                   selection_data_eem  = selection_eem,
+                   selection_mc_eem    = selection_eem + [cuts_eem.selections['is_prompt_lepton']],
+
+                   selection_tight = cuts_mmm.selections_pd['tight'],
+
+                   epochs = 20,
+                   )
+
+if __name__ == '__main__':
+    trainer.train()