epsilon-compositionality

commit fb0d5df815b6bf7e18d882800a63e92f6811495b
parent 347e99fd8c16b2dd6ea5971c79f076e2ef9f9a4f
Author: Étienne Simon <esimon@esimon.eu>
Date:   Mon,  6 Oct 2025 07:00:27 +0200

add dataset statistics

Diffstat:
M
epsilon_compositionality/compute_statistics.py
 | 
29
+++++++++++++++++++++++++++++

1 file changed, 29 insertions(+), 0 deletions(-)
diff --git a/epsilon_compositionality/compute_statistics.py b/epsilon_compositionality/compute_statistics.py
@@ -51,6 +51,34 @@ def read_similarities(epsilon_method: str, offset: int = 0) -> dict[str, Any]:
     return models
 
 
+def latex_data_statistics(output_path: pathlib.Path) -> None:
+    num_compounds: int = 0
+    num_sentences: int = 0
+    num_words: int = 0
+    num_epsilons: int = 0
+
+    model_path: pathlib.Path = next(INPUT_PATH.iterdir())
+    with model_path.open("r") as model_file:
+        for line in model_file:
+            compound = json.loads(line)
+            num_compounds += 1
+            for side in ["head", "modifier"]:
+                for sentence in compound[side]:
+                    num_sentences += 1
+                    num_words += len(sentence["positives"][0]["context_ac"].split())
+                    for polarity in ["positives", "negatives"]:
+                        num_epsilons += len(sentence[polarity])
+
+    num_sentences //= 2
+    num_words //= 2
+    with output_path.open("w") as output:
+        print(r"\NewDocumentCommand\dataNumCompounds{}{" + str(num_compounds) + "}", file=output)
+        print(r"\NewDocumentCommand\dataNumSentences{}{" + str(num_sentences) + "}", file=output)
+        print(r"\NewDocumentCommand\dataNumEpsilons{}{" + str(num_epsilons) + "}", file=output)
+        print(r"\NewDocumentCommand\dataAvgSentenceLength{}{" + f"{num_words / num_sentences:4.1f}" + "}", file=output)
+        print(r"\NewDocumentCommand\dataAvgAlternatives{}{" + f"{num_epsilons / num_sentences / 2:3.1f}" + "}", file=output)
+
+
 def table_cell(positives: list[float], negatives: list[float], method: str) -> str:
     wilcoxon = scipy.stats.wilcoxon(positives, negatives, alternative="greater")
     if method == "rank biserial":
@@ -117,6 +145,7 @@ def write_violins(output_path: pathlib.Path) -> None:
 def main() -> None:
     models = read_similarities("multiplicative")
     RESULTS_PATH.mkdir(exist_ok=True)
+    latex_data_statistics(RESULTS_PATH / "data statistics.tex")
     latex_main_statistics_table(RESULTS_PATH / "main statistics.tex", models)
     latex_distances_statistics(RESULTS_PATH / "distances statistics.tex")
     write_violins(RESULTS_PATH / "violins.dat")