Merge pull request #92 from uhh-cms/feature/cf_multi_config_task

MultiConfigPlotting and corresponding analysis changes

hbw/analysis/create_analysis.py

@@ -19,6 +19,12 @@
 logger = law.logger.get_logger(__name__)
 
 
+from hbw.config.defaults_and_groups import (
+    default_calibrator, default_selector, default_producers, default_ml_model,
+    ml_inputs_producer,
+)
+
+
 @timeit_multiple
 def create_hbw_analysis(
     name,
@@ -62,6 +68,13 @@ def create_hbw_analysis(
     # (used in wrapper_factory)
     analysis_inst.set_aux("config_groups", {})
 
+    analysis_inst.x.default_calibrator = default_calibrator(analysis_inst)
+    analysis_inst.x.default_selector = default_selector(analysis_inst)
+    analysis_inst.x.default_producer = default_producers
+    analysis_inst.x.default_weight_producer = "default"
+    analysis_inst.x.ml_inputs_producer = ml_inputs_producer(analysis_inst)
+    analysis_inst.x.default_ml_model = default_ml_model
+
     #
     # define configs
     #

hbw/categorization/categories.py

@@ -275,7 +275,7 @@ def catid_2b(self: Categorizer, events: ak.Array, **kwargs) -> tuple[ak.Array, a
     "hh_ggf_hbb_hvvqqlnu_kl1_kt1", "hh_vbf_hbb_hvvqqlnu_kv1_k2v1_kl1",
     "hh_ggf_hbb_hvv2l2nu_kl1_kt1", "hh_vbf_hbb_hvv2l2nu_kv1_k2v1_kl1",
     "tt", "st", "w_lnu", "dy", "v_lep", "h",
-    "dy_m50toinf",
+    "dy_m50toinf", "tt_dl",
 ]
 for proc in ml_processes:
     @categorizer(

hbw/config/defaults_and_groups.py

@@ -107,15 +107,16 @@ def set_config_defaults_and_groups(config_inst):
     # Defaults
     #
 
+    # NOTE: many of these have been moved to analysis_inst, TODO cleanup
     # TODO: the default dataset is currently still being set up by the law.cfg
     config_inst.x.default_dataset = default_signal_dataset = f"{default_signal_process}_{signal_generator}"
-    config_inst.x.default_calibrator = default_calibrator(config_inst)
-    config_inst.x.default_selector = default_selector(config_inst)
+    # config_inst.x.default_calibrator = default_calibrator(config_inst)
+    # config_inst.x.default_selector = default_selector(config_inst)
     config_inst.x.ml_inputs_producer = ml_inputs_producer(config_inst)
-    config_inst.x.default_producer = default_producers
-    config_inst.x.default_weight_producer = "default"
-    # config_inst.x.default_weight_producer = "btag_not_normalized"
-    config_inst.x.default_ml_model = default_ml_model
+    # config_inst.x.default_producer = default_producers
+    # config_inst.x.default_weight_producer = "default"
+    # # config_inst.x.default_weight_producer = "btag_not_normalized"
+    # config_inst.x.default_ml_model = default_ml_model
     config_inst.x.default_inference_model = "default" if year == 2017 else "sl_22"
     config_inst.x.default_categories = ["incl"]
     config_inst.x.default_variables = ["jet1_pt"]
@@ -205,6 +206,10 @@ def set_config_defaults_and_groups(config_inst):
         "dilep": ["tt_*", "st_*", "dy_*", "w_lnu_*", "hh_ggf_*"],
         "h": ["h_ggf_*", "h_vbf_*", "zh_*", "wph_*", "wmh_*", "tth_*", "ttzh_*", "ttwh_*"],
     }
+    if config_inst.name == "l22post":
+        config_inst.x.dataset_groups["test123"] = ["tt_dl_powheg", "tt_sl_powheg"]
+    elif config_inst.name == "l22pre":
+        config_inst.x.dataset_groups["test123"] = ["tt_dl_powheg"]
 
     # category groups for conveniently looping over certain categories
     # (used during plotting and for rebinning)

hbw/config/processes.py

@@ -11,6 +11,8 @@
 
 from cmsdb.util import add_decay_process
 
+from hbw.config.styling import color_palette
+
 
 def add_parent_process(config: od.Config, child_procs: list[od.Process], **kwargs):
     """
@@ -185,3 +187,15 @@ def configure_hbw_processes(config: od.Config):
                     xsecs=None,
                     aux={"flavour": flavour},
                 )
+
+    # create main background process
+    background = config.add_process(
+        name="background",
+        id=99999,
+        label="background",
+        color=color_palette["blue"],
+    )
+    for bg in ["tt", "dy", "st", "vv", "w_lnu", "h"]:
+        if config.has_process(bg):
+            bg = config.get_process(bg)
+            background.add_process(bg)

hbw/ml/base.py

@@ -196,31 +196,32 @@ def setup(self):
         # NOTE: since these variables are only used in ConfigTasks,
         #       we do not need to add these variables to all configs
         for proc in self.processes:
-            if f"mlscore.{proc}" not in self.config_inst.variables:
-                self.config_inst.add_variable(
-                    name=f"mlscore.{proc}",
-                    null_value=-1,
-                    binning=(1000, 0., 1.),
-                    x_title=f"DNN output score {self.config_inst.get_process(proc).x.ml_label}",
-                    aux={"rebin": 25},  # automatically rebin to 40 bins for plotting tasks
-                )
-
-    def preparation_producer(self: MLModel, config_inst: od.Config):
+            for config_inst in self.config_insts:
+                if f"mlscore.{proc}" not in config_inst.variables:
+                    config_inst.add_variable(
+                        name=f"mlscore.{proc}",
+                        null_value=-1,
+                        binning=(1000, 0., 1.),
+                        x_title=f"DNN output score {config_inst.get_process(proc).x.ml_label}",
+                        aux={"rebin": 25},  # automatically rebin to 40 bins for plotting tasks
+                    )
+
+    def preparation_producer(self: MLModel, analysis_inst: od.Analysis):
         """ producer that is run as part of PrepareMLEvents and MLEvaluation (before `evaluate`) """
         return "ml_preparation"
 
-    def training_calibrators(self, config_inst: od.Config, requested_calibrators: Sequence[str]) -> list[str]:
+    def training_calibrators(self, analysis_inst: od.Analysis, requested_calibrators: Sequence[str]) -> list[str]:
         # fix MLTraining Phase Space
         # NOTE: since automatic resolving is not working here, we do it ourselves
-        return requested_calibrators or [config_inst.x.default_calibrator]
+        return requested_calibrators or [analysis_inst.x.default_calibrator]
 
-    def training_producers(self, config_inst: od.Config, requested_producers: Sequence[str]) -> list[str]:
+    def training_producers(self, analysis_inst: od.Analysis, requested_producers: Sequence[str]) -> list[str]:
         # fix MLTraining Phase Space
         # NOTE: might be nice to keep the "pre_ml_cats" for consistency, but running two
         # categorization Producers in the same workflow is messy, so we skip it for now
-        # return requested_producers or ["event_weights", "pre_ml_cats", config_inst.x.ml_inputs_producer]
-        # return requested_producers or ["event_weights", config_inst.x.ml_inputs_producer]
-        return ["event_weights", config_inst.x.ml_inputs_producer]
+        # return requested_producers or ["event_weights", "pre_ml_cats", analysis_inst.x.ml_inputs_producer]
+        # return requested_producers or ["event_weights", analysis_inst.x.ml_inputs_producer]
+        return ["event_weights", analysis_inst.x.ml_inputs_producer]
 
     def requires(self, task: law.Task) -> dict[str, Any]:
         # Custom requirements (none currently)

hbw/ml/derived/dl.py

@@ -136,22 +136,23 @@ def setup(self):
         # NOTE: since these variables are only used in ConfigTasks,
         #       we do not need to add these variables to all configs
         for proc in self.processes:
-            if f"mlscore.{proc}_manybins" not in self.config_inst.variables:
-                self.config_inst.add_variable(
-                    name=f"mlscore.{proc}_manybins",
-                    expression=f"mlscore.{proc}",
-                    null_value=-1,
-                    binning=(1000, 0., 1.),
-                    x_title=f"DNN output score {self.config_inst.get_process(proc).x.ml_label}",
-                    aux={"rebin": 40},
-                )
-                self.config_inst.add_variable(
-                    name=f"mlscore40.{proc}",
-                    expression=f"mlscore.{proc}",
-                    null_value=-1,
-                    binning=(40, 0., 1.),
-                    x_title=f"DNN output score {self.config_inst.get_process(proc).x.ml_label}",
-                )
+            for config_inst in self.config_insts:
+                if f"mlscore.{proc}_manybins" not in config_inst.variables:
+                    config_inst.add_variable(
+                        name=f"mlscore.{proc}_manybins",
+                        expression=f"mlscore.{proc}",
+                        null_value=-1,
+                        binning=(1000, 0., 1.),
+                        x_title=f"DNN output score {config_inst.get_process(proc).x.ml_label}",
+                        aux={"rebin": 40},
+                    )
+                    config_inst.add_variable(
+                        name=f"mlscore40.{proc}",
+                        expression=f"mlscore.{proc}",
+                        null_value=-1,
+                        binning=(40, 0., 1.),
+                        x_title=f"DNN output score {config_inst.get_process(proc).x.ml_label}",
+                    )
 
 
 #
@@ -271,5 +272,6 @@ def setup(self):
 dl_22.derive("dl_22_v1")
 dl_22_limited = dl_22post.derive("dl_22_limited", cls_dict={
     "training_configs": lambda self, requested_configs: ["l22pre", "l22post"],
-    "processes": ["hh_ggf_hbb_hvv2l2nu_kl1_kt1", "st_tchannel_t"],
+    "epochs": 4,
+    "processes": ["hh_ggf_hbb_hvv2l2nu_kl1_kt1", "tt_dl"],
 })

hbw/tasks/corrections.py

@@ -6,7 +6,7 @@
 
 from functools import cached_property
 
-from columnflow.tasks.framework.base import Requirements
+from columnflow.tasks.framework.base import Requirements, ConfigTask
 from columnflow.tasks.framework.mixins import (
     SelectorMixin, CalibratorsMixin,
 )
@@ -25,6 +25,7 @@
 
 class GetBtagNormalizationSF(
     HBWTask,
+    ConfigTask,
     SelectorMixin,
     CalibratorsMixin,
     # law.LocalWorkflow,

hbw/tasks/ml.py

@@ -22,7 +22,6 @@
     CalibratorsMixin,
     ProducersMixin,
     MLModelTrainingMixin,
-    MLModelMixin,
     MLModelsMixin,
     MLModelDataMixin,
     SelectorStepsMixin,
@@ -224,39 +223,23 @@ def workflow_requires(self):
                     self,
                     config=config_inst.name,
                     dataset=dataset_inst.name,
-                    calibrators=_calibrators,
-                    selector=_selector,
-                    producers=_producers,
                     branch=-1,
                 )
                 for dataset_inst in dataset_insts
             }
-            for (config_inst, dataset_insts), _calibrators, _selector, _producers in zip(
-                self.ml_model_inst.used_datasets.items(),
-                self.calibrators,
-                self.selectors,
-                self.producers,
-            )
+            for config_inst, dataset_insts in self.ml_model_inst.used_datasets.items()
         }
         reqs["stats"] = {
             config_inst.name: {
                 dataset_inst.name: self.reqs.MergeMLStats.req(
                     self,
                     config=config_inst.name,
                     dataset=dataset_inst.name,
-                    calibrators=_calibrators,
-                    selector=_selector,
-                    producers=_producers,
                     tree_index=-1,
                 )
                 for dataset_inst in dataset_insts
             }
-            for (config_inst, dataset_insts), _calibrators, _selector, _producers in zip(
-                self.ml_model_inst.used_datasets.items(),
-                self.calibrators,
-                self.selectors,
-                self.producers,
-            )
+            for config_inst, dataset_insts in self.ml_model_inst.used_datasets.items()
         }
         return reqs
 
@@ -275,20 +258,12 @@ def requires(self):
                     self,
                     config=config_inst.name,
                     dataset=dataset_inst.name,
-                    calibrators=_calibrators,
-                    selector=_selector,
-                    producers=_producers,
                     branch=-1,
                 )
                 for dataset_inst in dataset_insts
                 if dataset_inst.x.ml_process == process
             }
-            for (config_inst, dataset_insts), _calibrators, _selector, _producers in zip(
-                self.ml_model_inst.used_datasets.items(),
-                self.calibrators,
-                self.selectors,
-                self.producers,
-            )
+            for config_inst, dataset_insts in self.ml_model_inst.used_datasets.items()
         }
 
         # load stats for all processes
@@ -298,18 +273,10 @@ def requires(self):
                     self,
                     config=config_inst.name,
                     dataset=dataset_inst.name,
-                    calibrators=_calibrators,
-                    selector=_selector,
-                    producers=_producers,
                     tree_index=-1)
                 for dataset_inst in dataset_insts
             }
-            for (config_inst, dataset_insts), _calibrators, _selector, _producers in zip(
-                self.ml_model_inst.used_datasets.items(),
-                self.calibrators,
-                self.selectors,
-                self.producers,
-            )
+            for config_inst, dataset_insts in self.ml_model_inst.used_datasets.items()
         }
 
         return reqs
@@ -435,13 +402,9 @@ def workflow_run(self):
 
 
 class MLEvaluationSingleFold(
-    # NOTE: this should probably be a MLModelTrainingMixin, but I'll postpone this until the MultiConfigTask
-    # is implemented
+    # NOTE: mixins might need fixing, needs to be checked
     HBWTask,
-    MLModelMixin,
-    ProducersMixin,
-    SelectorMixin,
-    CalibratorsMixin,
+    MLModelTrainingMixin,
     law.LocalWorkflow,
     RemoteWorkflow,
 ):
@@ -489,16 +452,10 @@ def workflow_requires(self):
             self,
             branches=(self.fold,),
             configs=(self.config_inst.name,),
-            calibrators=(self.calibrators,),
-            selectors=(self.selector,),
-            producers=(self.producers,),
         )
         reqs["preml"] = self.reqs.MLPreTraining.req_different_branching(
             self,
             configs=(self.config_inst.name,),
-            calibrators=(self.calibrators,),
-            selectors=(self.selector,),
-            producers=(self.producers,),
             branch=-1,
         )
         # reqs["preml"] = self.reqs.MLPreTraining.req_different_branching(self, branch=-1)
@@ -515,16 +472,10 @@ def requires(self):
             branches=(self.fold,),
             branch=self.fold,
             configs=(self.config_inst.name,),
-            calibrators=(self.calibrators,),
-            selectors=(self.selector,),
-            producers=(self.producers,),
         )
         reqs["preml"] = self.reqs.MLPreTraining.req_different_branching(
             self,
             configs=(self.config_inst.name,),
-            calibrators=(self.calibrators,),
-            selectors=(self.selector,),
-            producers=(self.producers,),
             branch=-1,
         )
         # reqs["preml"] = self.reqs.MLPreTraining.req_different_branching(self, branch=-1)
@@ -585,13 +536,9 @@ def run(self):
 
 
 class PlotMLResultsSingleFold(
-    # NOTE: this should probably be a MLModelTrainingMixin, but I'll postpone this until the MultiConfigTask
-    # is implemented
+    # NOTE: mixins might need fixing, needs to be checked
     HBWTask,
-    MLModelMixin,
-    ProducersMixin,
-    SelectorMixin,
-    CalibratorsMixin,
+    MLModelTrainingMixin,
     law.LocalWorkflow,
     RemoteWorkflow,
 ):
@@ -649,19 +596,13 @@ def requires(self):
             "training": self.reqs.MLTraining.req_different_branching(
                 self,
                 configs=(self.config_inst.name,),
-                calibrators=(self.calibrators,),
-                selectors=(self.selector,),
-                producers=(self.producers,),
                 branches=(self.fold,),
             ),
         }
 
         reqs["preml"] = self.reqs.MLPreTraining.req_different_branching(
             self,
             configs=(self.config_inst.name,),
-            calibrators=(self.calibrators,),
-            selectors=(self.selector,),
-            producers=(self.producers,),
             branch=-1,
         )
         reqs["mlpred"] = {