fix: take filter_expr into account in dependency parsing evaluation

Author: percevalw

changelog.md

@@ -15,6 +15,7 @@
 - Support packaging with poetry 2.0
 - Solve pickling issues with multiprocessing when pytorch is installed
 - Allow deep attributes like `a.b.c` for `span_attributes` in Standoff and OMOP doc2dict converters
+- Take `filter_expr` into account in dependency parsing evaluation
 
 # v0.15.0 (2024-12-13)
 

edsnlp/metrics/dep_parsing.py

@@ -33,6 +33,9 @@ def dependency_parsing_metric(
         examples = [eg for eg in examples if filter_fn(eg.reference)]
 
     for eg_idx, eg in enumerate(examples):
+        if filter_expr and not eval(filter_expr, {}, eg):
+            continue
+
         for token in eg.reference:
             items["uas"][0].add((eg_idx, token.i, token.head.i))
             items["las"][0].add((eg_idx, token.i, token.head.i, token.dep_))

tests/training/dep_parser_config.yml

@@ -23,6 +23,7 @@ scorer:
   speed: false
   dep:
     '@metrics': "eds.dep_parsing"
+    filter_expr: "doc.text != ''"
 
 # 🎛️ OPTIMIZER
 optimizer: