make threshold for parameter relevance heuristic configurable

1 files changed, 12 insertions, 3 deletions

diff --git a/lib/parameters.py b/lib/parameters.py
index a3c5ab1..74b9091 100644
--- a/lib/parameters.py
+++ b/lib/parameters.py
@@ -42,7 +42,7 @@ def distinct_param_values(param_tuples):
     return distinct_values
 
 
-def _depends_on_param(corr_param, std_param, std_lut):
+def _depends_on_param(corr_param, std_param, std_lut, threshold=0.5):
     # if self.use_corrcoef:
     if False:
         return corr_param > 0.1
@@ -50,7 +50,7 @@ def _depends_on_param(corr_param, std_param, std_lut):
         # In general, std_param_lut < std_by_param. So, if std_by_param == 0, std_param_lut == 0 follows.
         # This means that the variation of param does not affect the model quality -> no influence
         return False
-    return std_lut / std_param < 0.5
+    return std_lut / std_param < threshold
 
 
 def _mean_std_by_param(n_by_param, all_param_values, param_index):
@@ -190,6 +190,8 @@ def _compute_param_statistics(
 
     np.seterr("raise")
 
+    relevance_threshold = float(os.getenv("DFATOOL_PARAM_RELEVANCE_TRESHOLD", 0.5))
+
     for param_idx, param in enumerate(param_names):
         if param_idx < len(codependent_params) and codependent_params[param_idx]:
             by_param = partition_by_param(
@@ -209,6 +211,7 @@ def _compute_param_statistics(
             ret["corr_by_param"][param],
             ret["std_by_param"][param],
             ret["std_param_lut"],
+            relevance_threshold,
         )
 
     if arg_count:
@@ -1124,6 +1127,8 @@ class ModelAttribute:
                 "build_dtree {self.name} {self.attr} called with loss_ignore_scalar=True, with_function_leaves=False. This does not make sense."
             )
 
+        relevance_threshold = float(os.getenv("DFATOOL_PARAM_RELEVANCE_TRESHOLD", 0.5))
+
         self.model_function = self._build_dtree(
             parameters,
             data,
@@ -1132,6 +1137,7 @@ class ModelAttribute:
             ignore_irrelevant_parameters=ignore_irrelevant_parameters,
             loss_ignore_scalar=loss_ignore_scalar,
             threshold=threshold,
+            relevance_threshold=relevance_threshold,
         )
 
     def _build_dtree(
@@ -1143,6 +1149,7 @@ class ModelAttribute:
         ignore_irrelevant_parameters=True,
         loss_ignore_scalar=False,
         threshold=100,
+        relevance_threshold=0.5,
         level=0,
     ):
         """
@@ -1206,7 +1213,9 @@ class ModelAttribute:
                 std_by_param = _mean_std_by_param(
                     by_param, distinct_values_by_param_index, param_index
                 )
-                if not _depends_on_param(None, std_by_param, std_lut):
+                if not _depends_on_param(
+                    None, std_by_param, std_lut, relevance_threshold
+                ):
                     loss.append(np.inf)
                     continue