epsilon-compositionality

commit c82e284c00e85275181f14259dda076fd844b902
parent 6dfb0f1474a4e155093ea374865e5005c3e22bfe
Author: Sondre Wold <sondrewold1@gmail.com>
Date:   Wed, 24 Sep 2025 16:43:04 +0200

add class to results

Diffstat:
M
experiment.py
 | 
13
++++++++++---
M
get_epsilon.py
 | 
3
+++

2 files changed, 13 insertions(+), 3 deletions(-)
diff --git a/experiment.py b/experiment.py
@@ -52,6 +52,7 @@ class Point(NamedTuple):
     phrase_b_embedded: torch.Tensor
     context_ac_embedded: torch.Tensor
     context_bc_embedded: torch.Tensor
+    comp_class: str
 
 
 class ResultObject(NamedTuple):
@@ -63,6 +64,7 @@ class ResultObject(NamedTuple):
 class SentencePairs(NamedTuple):
     sentence_A: list[str]
     sentence_B: list[str]
+    classes: list[str]
 
 
 def normalize_v2():
@@ -543,7 +545,7 @@ class Experiment:
         self.phrase_b_length: list[int] = []
 
     def run_experiment(self) -> None:
-        for i, (sentence_a, sentence_b) in enumerate(tqdm.tqdm(zip(*self.sentence_pairs))):
+        for i, (sentence_a, sentence_b, cls) in enumerate(tqdm.tqdm(zip(*self.sentence_pairs))):
             diffs: list[str] | None = find_diff(sentence_a, sentence_b)
             if not diffs:
                 self.diff_errors += 1
@@ -611,6 +613,7 @@ class Experiment:
                     phrase_b_embedded=phrase_b_embedded,
                     context_ac_embedded=a_in_context_embedded,
                     context_bc_embedded=b_in_context_embedded,
+                    comp_class=cls
                 )
             )
 
@@ -647,7 +650,7 @@ def main() -> None:
         synonym_sentence: list[str] = df[
             "synonym both"
         ].values.tolist()  # Both the modifier and the noun are swapped with synonyms from word net.
-
+        
         sentence_A: list[str] = original_sentence
         sentence_B: list[str] = synonym_sentence
 
@@ -675,7 +678,11 @@ def main() -> None:
     else:
         raise ValueError("Undefined dataset type...")
 
-    sentence_pairs = SentencePairs(sentence_A=sentence_A, sentence_B=sentence_B)
+    clss: list[str] = df[
+            "class"
+        ].values.tolist()  # The comp. classes
+
+    sentence_pairs = SentencePairs(sentence_A=sentence_A, sentence_B=sentence_B, classes=clss)
 
     models = (
         [args.model] if args.model is not None else list(ModelFactory._factory.keys())
diff --git a/get_epsilon.py b/get_epsilon.py
@@ -91,6 +91,7 @@ def main() -> None:
         "divisive": [],
         "additive": [],
         "simscores": [],
+        "classes": [],
     }
     for raw_file in raw_files:
         with open(raw_file, "rb") as file:
@@ -104,6 +105,7 @@ def main() -> None:
             context_sims: numpy.ndarray = numpy.array(
                 [point.contextualized_similarity for point in points]
             )
+            classes = [point.comp_class for point in points]
             ac_embedded: list[torch.Tensor] = [torch.Tensor(point.context_ac_embedded) for point in points]
             bc_embedded: list[torch.Tensor] = [torch.Tensor(point.context_bc_embedded) for point in points]
             tmp_result: dict[str, list[float]] = get_epsilons(word_sims, context_sims)
@@ -131,6 +133,7 @@ def main() -> None:
             final_result["original"].append(final_original)
             final_result["divisive"].append(final_division)
             final_result["additive"].append(final_additive)
+            final_result["classes"] = classes
 
     # df: pandas.DataFrame = pandas.DataFrame.from_dict(final_result)
     # df.to_csv(args.result_path, sep=",", index=False)