commit 57a8735d2128141ede3e7b5ccb176e37aa8ec864
parent ba126b546e384dedfaa083c1b48f227bae104fa3
Author: Sondre Wold <sondrewold1@gmail.com>
Date: Wed, 27 Aug 2025 14:32:11 +0200
scripts for debugging
Diffstat:
2 files changed, 201 insertions(+), 0 deletions(-)
diff --git a/debug/analyze_embeddings.py b/debug/analyze_embeddings.py
@@ -0,0 +1,114 @@
+import argparse
+import pathlib
+import pickle
+
+from numpy import mean
+from experiment import Point
+
+import pandas as pd
+
+from typing import Any
+
+import torch
+from get_epsilon import get_dataset_and_model_name
+
+
+def parse_args() -> argparse.Namespace:
+ parser = argparse.ArgumentParser(description="Gather epsilon values")
+ parser.add_argument("-points_path", type=pathlib.Path)
+ parser.add_argument("-result_path", type=pathlib.Path)
+ return parser.parse_args()
+
+
+def main() -> None:
+ args = parse_args()
+ points_folder: str = args.points_path
+ raw_files = list(pathlib.Path(points_folder).glob("*.pkl"))
+
+ final_result: dict[str, list[Any]] = {
+ "dataset": [],
+ "model": [],
+ "a_norm": [],
+ "b_norm": [],
+ "ac_norm": [],
+ "bc_norm": [],
+ "a_mean": [],
+ "b_mean": [],
+ "ac_mean": [],
+ "bc_mean": [],
+ "distance": [],
+ }
+
+ for raw_file in raw_files:
+ with open(raw_file, "rb") as file:
+ points: list[Point] = pickle.load(file)
+ dataset_name, model_name = get_dataset_and_model_name(raw_file.name)
+ a_norms: list[float] = []
+ b_norms: list[float] = []
+ ac_norms: list[float] = []
+ bc_norms: list[float] = []
+ a_means: list[torch.Tensor] = []
+ b_means: list[torch.Tensor] = []
+ ac_means: list[torch.Tensor] = []
+ bc_means: list[torch.Tensor] = []
+ for point in points:
+ a_norms.append(
+ torch.linalg.norm(torch.Tensor(point.phrase_a_embedded)).item()
+ )
+ b_norms.append(
+ torch.linalg.norm(torch.Tensor(point.phrase_b_embedded)).item()
+ )
+ ac_norms.append(
+ torch.linalg.norm(torch.Tensor(point.context_ac_embedded)).item()
+ )
+ bc_norms.append(
+ torch.linalg.norm(torch.Tensor(point.context_bc_embedded)).item()
+ )
+ a_means.append(torch.Tensor(point.phrase_a_embedded))
+ b_means.append(torch.Tensor(point.phrase_b_embedded))
+ ac_means.append(torch.Tensor(point.context_ac_embedded))
+ bc_means.append(torch.Tensor(point.context_bc_embedded))
+
+ final_result["dataset"].append(dataset_name)
+ final_result["model"].append(model_name)
+ final_result["a_norm"].append(mean(a_norms))
+ final_result["b_norm"].append(mean(b_norms))
+ final_result["ac_norm"].append(mean(ac_norms))
+ final_result["bc_norm"].append(mean(bc_norms))
+
+ final_result["a_mean"].append(
+ torch.mean(torch.stack(a_means), dim=-1).norm().item()
+ )
+ final_result["b_mean"].append(
+ torch.mean(torch.stack(b_means), dim=-1).norm().item()
+ )
+ final_result["ac_mean"].append(
+ torch.mean(torch.stack(ac_means), dim=-1).norm().item()
+ )
+ final_result["bc_mean"].append(
+ torch.mean(torch.stack(bc_means), dim=-1).norm().item()
+ )
+
+ final_result["distance"].append(
+ (
+ (
+ torch.mean(torch.stack(ac_means), dim=-1)
+ + torch.mean(torch.stack(bc_means), dim=-1)
+ )
+ / 2
+ - (
+ torch.mean(torch.stack(a_means), dim=-1)
+ + torch.mean(torch.stack(b_means), dim=-1)
+ )
+ / 2
+ )
+ .norm()
+ .item()
+ )
+
+ df = pd.DataFrame.from_dict(final_result)
+ print(df)
+
+
+if __name__ == "__main__":
+ main()
diff --git a/debug/plot_sims.py b/debug/plot_sims.py
@@ -0,0 +1,87 @@
+import argparse
+import pathlib
+import pickle
+
+import numpy
+
+from numpy import mean
+from experiment import Point
+
+import matplotlib.pyplot as plt
+
+import pandas as pd
+
+from typing import Any
+
+import torch
+from get_epsilon import get_dataset_and_model_name
+
+
+def parse_args() -> argparse.Namespace:
+ parser = argparse.ArgumentParser(description="Gather epsilon values")
+ parser.add_argument("-points_path", type=pathlib.Path)
+ parser.add_argument("-result_path", type=pathlib.Path)
+ return parser.parse_args()
+
+
+def main() -> None:
+ args = parse_args()
+ points_folder: str = args.points_path
+ raw_files = list(pathlib.Path(points_folder).glob("*.pkl"))
+
+ final_result: dict[str, list[Any]] = {
+ "dataset": [],
+ "model": [],
+ "phrase_sim": [],
+ "context_sim": [],
+ }
+
+ for raw_file in raw_files:
+ with open(raw_file, "rb") as file:
+ points: list[Point] = pickle.load(file)
+ dataset_name, model_name = get_dataset_and_model_name(raw_file.name)
+
+ word_sims = [
+ (
+ point.phrase_similarity
+ if isinstance(point.phrase_similarity, float)
+ else point.phrase_similarity.item()
+ )
+ for point in points
+ ]
+ context_sims = [
+ (
+ point.contextualized_similarity
+ if isinstance(point.contextualized_similarity, float)
+ else point.contextualized_similarity.item()
+ )
+ for point in points
+ ]
+
+ final_result["dataset"].append(dataset_name)
+ final_result["model"].append(model_name)
+ final_result["phrase_sim"].append(word_sims)
+ final_result["context_sim"].append(context_sims)
+
+ fig, axs = plt.subplots(nrows=2, ncols=2, figsize=(10, 4))
+ axs[0, 0].violinplot(
+ final_result["phrase_sim"],
+ [y + 1 for y in range(len(final_result["phrase_sim"]))],
+ )
+ axs[0, 0].set_xticks(
+ [y + 1 for y in range(len(final_result["phrase_sim"]))],
+ labels=final_result["model"],
+ )
+ axs[1, 0].violinplot(
+ final_result["context_sim"],
+ [y + 1 for y in range(len(final_result["context_sim"]))],
+ )
+ axs[1, 0].set_xticks(
+ [y + 1 for y in range(len(final_result["context_sim"]))],
+ labels=final_result["model"],
+ )
+ plt.show()
+
+
+if __name__ == "__main__":
+ main()
