update MVA models

configs/dqcd/models.yml

@@ -95,11 +95,11 @@ xgb01:
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
-    booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
+    num_boost_round: 90       # number of epochs (equal to the number of trees!)
+    booster: 'gbtree'         # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
-    device:      'auto'      # 'auto', 'cpu', 'cuda'
-
+    device:      'auto'       # 'auto', 'cpu', 'cuda'
+    
     learning_rate: 0.1
     gamma: 1.67
     max_depth: 10
@@ -139,7 +139,7 @@ xgb01-no-mass:
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
+    num_boost_round: 90      # number of epochs (equal to the number of trees!)
     booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
     device:      'auto'      # 'auto', 'cpu', 'cuda'
@@ -171,10 +171,10 @@ xgb01-no-mass:
 
 # XGBoost
 # https://xgboost.readthedocs.io/en/latest/parameter.html
-xgb01-only-mu:
+xgb01-no-jets:
   train:   'xgb'
   predict: 'xgb'
-  label:   'XGB-only-mu'
+  label:   'XGB-no-jets'
   raytune:  xgb_trial_0
 
   # ** Custom set of variables **
@@ -183,7 +183,7 @@ xgb01-only-mu:
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
+    num_boost_round: 90      # number of epochs (equal to the number of trees!)
     booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
     device:      'auto'      # 'auto', 'cpu', 'cuda'
@@ -217,16 +217,16 @@ xgb01-only-mu:
 # https://xgboost.readthedocs.io/en/latest/parameter.html
 iceboost:
   train:   'xgb'
-  predict: 'xgb'
+  predict: 'xgb_logistic'     # We apply logistic link function manually (because custom loss)
   label:   'ICEBOOST'
   raytune:  xgb_trial_0
-
+  
   #exclude_MVA_vars: ['.*']
   #include_MVA_vars: ['.*']
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
+    num_boost_round: 90      # number of epochs (equal to the number of trees!)
     booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
     device:      'auto'      # 'auto', 'cpu', 'cuda'
@@ -265,7 +265,7 @@ iceboost:
 # https://xgboost.readthedocs.io/en/latest/parameter.html
 iceboost-no-mass:
   train:   'xgb'
-  predict: 'xgb'
+  predict: 'xgb_logistic'     # We apply logistic link function manually (because custom loss)
   label:   'ICEBOOST-no-mass'
   raytune:  xgb_trial_0
 
@@ -275,7 +275,7 @@ iceboost-no-mass:
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
+    num_boost_round: 90      # number of epochs (equal to the number of trees!)
     booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
     device:      'auto'      # 'auto', 'cpu', 'cuda'
@@ -312,10 +312,10 @@ iceboost-no-mass:
 
 # XGBoost
 # https://xgboost.readthedocs.io/en/latest/parameter.html
-iceboost-only-mu:
+iceboost-no-jets:
   train:   'xgb'
-  predict: 'xgb'
-  label:   'ICEBOOST-only-mu'
+  predict: 'xgb_logistic'     # We apply logistic link function manually (because custom loss)
+  label:   'ICEBOOST-no-jets'
   raytune:  xgb_trial_0
 
   # ** Custom set of variables **
@@ -324,7 +324,7 @@ iceboost-only-mu:
 
   # general parameters
   model_param:
-    num_boost_round: 100      # number of epochs (equal to the number of trees!)
+    num_boost_round: 90      # number of epochs (equal to the number of trees!)
     booster: 'gbtree'        # 'gbtree' (default), 'dart' (dropout boosting)
     tree_method: 'hist'
     device:      'auto'      # 'auto', 'cpu', 'cuda'

configs/dqcd/mvavars.py

@@ -266,12 +266,12 @@
 
 # Mutual information regularization targets
 MI_VARS = [
-  'muonSV_mass.*'
+  #'muonSV_mass.*'  # All
+  'muonSV_mass(0)'  # The leading one
 ]
 
 print(LOAD_VARS)
 
 # (regular expressions supported here)
 #LOAD_VARS = ['.+hlt.?', '.?gen.?']
 #LOAD_VARS = ['.*'] # all
-

configs/dqcd/plots.yml

@@ -33,7 +33,7 @@ FILTER: &FILTER
 
 
 basic:
-  active: False
+  active: True
   nbins:  70
   exclude_vals: [null, -999.0]
 
@@ -45,14 +45,14 @@ contours:
 
 ROC:
   active: true
-  num_bootstrap: 200
+  num_bootstrap: 50
   xmin: 1.0E-6
   set_filter: *FILTER
 
 ## Binned ROC plots can be 1D or 2D
 ROC_binned:
   active: false
-  num_bootstrap: 200
+  num_bootstrap: 50
   xmin: 1.0E-4
 
   #plot[0]:

configs/dqcd/plots_new.yml

@@ -33,7 +33,7 @@ FILTER: &FILTER
 
 
 basic:
-  active: False
+  active: True
   nbins:  70
   exclude_vals: [null, -999.0]
 
@@ -45,14 +45,14 @@ contours:
 
 ROC:
   active: true
-  num_bootstrap: 200
+  num_bootstrap: 50
   xmin: 1.0E-6
   set_filter: *FILTER
 
 ## Binned ROC plots can be 1D or 2D
 ROC_binned:
   active: false
-  num_bootstrap: 200
+  num_bootstrap: 50
   xmin: 1.0E-4
 
   #plot[0]:

configs/dqcd/tune0.yml

@@ -146,12 +146,12 @@ train_runmode:
 
     - xgb01
     - xgb01-no-mass
-    - xgb01-only-mu
+    - xgb01-no-jets
 
     - iceboost
     - iceboost-no-mass
-    - iceboost-only-mu
-
+    - iceboost-no-jets
+    
     #- dmlp
     #- lgr0
 

configs/dqcd/tune0_new.yml

@@ -131,7 +131,7 @@ train_runmode:
     algo: 'truncate'   # algorithm: 'truncate', null
     qmin: 0.01         # in [0,100] 
     qmax: 99.9         # in [0,100]
-
+  
 
   # ** Activate models here **
   # Give all models some unique identifier and label
@@ -144,11 +144,11 @@ train_runmode:
 
     - xgb01
     - xgb01-no-mass
-    - xgb01-only-mu
+    - xgb01-no-jets
 
     - iceboost
     - iceboost-no-mass
-    - iceboost-only-mu
+    - iceboost-no-jets
 
     #- dmlp
     #- lgr0

icenet/tools/aux.py

@@ -8,6 +8,7 @@
 import awkward as ak
 import re
 import time
+import datetime
 
 import numba
 import copy
@@ -827,21 +828,34 @@ def deltar(eta1,eta2, phi1,phi2):
     return np.sqrt((eta1 - eta2)**2 + deltaphi(phi1,phi2)**2)
 
 
-def create_model_filename(path, label, epoch, filetype, max_epochs=int(1e5)):
+def getmtime(filename):
+    """ Return the last modification time of a file, reported by os.stat()
+    """
+    return os.stat(filename).st_mtime
 
-    def createfilename(i):
-        return path + '/' + label + '_' + str(i) + filetype
 
-    # Loop over epochs
-    last_found  = -1
-    for i in range(max_epochs):
-        filename = createfilename(i)
-        if os.path.exists(filename):
-            last_found = i
+def create_model_filename(path: str, label: str, filetype='.dat', epoch:int=None, max_epochs:int=int(1e5)):
+    """ Create model filename 
+    """
+    def createfilename(i):
+        return f'{path}/{label}_{i}{filetype}'
 
+    if epoch is None or epoch == -1:
+
+        last_found  = -1
+        for i in range(max_epochs):
+            filename = createfilename(i)
+            if os.path.exists(filename):
+                last_found = i
+            else:
+                break
+    else:
+        last_found = epoch
+
     filename = createfilename(last_found)
-    print(__name__ + f'.create_model_filename: Found model file: {filename}')
-
+    dt = datetime.datetime.fromtimestamp(getmtime(filename))
+    print(__name__ + f'.create_model_filename: Found a model file: {filename} (modified {dt})')
+
     return filename