Merge pull request #141 from HealthyPear/fix-energy_scale_in_and_from…

…_model

Ensure that estimated energy is always recorded in linear scale

protopipe/aux/example_config_files/AdaBoostRegressor.yaml

@@ -26,6 +26,7 @@ GridSearchCV:
 Method:
   name: 'sklearn.ensemble.AdaBoostRegressor'
   target_name: 'true_energy'
+  log_10_target: True # this makes the model use log10(target_name)
   # Please, see scikit-learn's API for what each parameter means
   # NOTE: null == None
   base_estimator:

protopipe/aux/example_config_files/RandomForestRegressor.yaml

@@ -26,7 +26,8 @@ GridSearchCV:
 # Definition of the model algorithm/method used and its hyper-parameters
 Method:
   name: 'sklearn.ensemble.RandomForestRegressor' # DO NOT CHANGE
-  target_name: 'log10_true_energy'
+  target_name: 'true_energy'
+  log_10_target: True # this makes the model use log10(target_name)
   tuned_parameters:
     # Please, see scikit-learn's API for what each parameter means
     # NOTE: null == None

protopipe/scripts/build_model.py

@@ -123,6 +123,14 @@ def main():
 
     if class_name in model_types["regressor"]:
 
+        try:
+            log_10_target = cfg["Method"]["log_10_target"]
+        except KeyError:
+            log_10_target = True
+
+        if log_10_target:
+            target_name = f"log10_{target_name}"
+
         # Get the selection cuts
         cuts = make_cut_list(cfg["SigFiducialCuts"])
 

protopipe/scripts/data_training.py

@@ -119,6 +119,7 @@ def main():
 
         # Read configuration file
         regressor_config = load_config(args.regressor_config)
+        log_10_target = regressor_config["Method"]["log_10_target"]
 
         regressor_files = (
             args.regressor_dir + "/regressor_{cam_id}_{regressor}.pkl.gz"
@@ -380,6 +381,10 @@ def main():
                         energy_tel[idx] = model.predict(features_values)
                         weight_tel[idx] = data["estimation_weight"]
 
+                    if log_10_target:
+                        energy_tel[idx] = 10**energy_tel[idx]
+                        weight_tel[idx] = 10**weight_tel[idx]
+
                     reco_energy_tel[tel_id] = energy_tel[idx]
 
                 reco_energy = np.sum(weight_tel * energy_tel) / sum(weight_tel)

protopipe/scripts/tests/test_AdaBoostRegressor.yaml

@@ -28,6 +28,7 @@ GridSearchCV:
 Method:
   name: 'sklearn.ensemble.AdaBoostRegressor'
   target_name: 'true_energy'
+  log_10_target: True # this makes the model use log10(target_name)
   # Please, see scikit-learn's API for what each parameter means
   # NOTE: null == None
   base_estimator:

protopipe/scripts/tests/test_RandomForestRegressor.yaml

@@ -27,7 +27,8 @@ GridSearchCV:
 # Definition of the model algorithm/method used and its hyper-parameters
 Method:
   name: 'sklearn.ensemble.RandomForestRegressor' # DO NOT CHANGE
-  target_name: 'log10_true_energy'
+  target_name: 'true_energy'
+  log_10_target: True # this makes the model use log10(target_name)
   tuned_parameters:
     # Please, see scikit-learn's API for what each parameter means
     # NOTE: null == None

protopipe/scripts/write_dl2.py

@@ -183,6 +183,7 @@ def main():
 
         # Read configuration file
         regressor_config = load_config(args.regressor_config)
+        log_10_target = regressor_config["Method"]["log_10_target"]
 
         regressor_files = (
             args.regressor_dir + "/regressor_{cam_id}_{regressor}.pkl.gz"
@@ -427,6 +428,10 @@ class RecoEvent(tb.IsDescription):
                     else:
                         energy_tel[idx] = np.nan
 
+                    if log_10_target:
+                        energy_tel[idx] = 10**energy_tel[idx]
+                        weight_tel[idx] = 10**weight_tel[idx]
+
                     # Record the values regardless of the validity
                     # We don't use this now, but it should be recorded
                     energy_tel_classifier[tel_id] = energy_tel[idx]