Feature/evaluation correlation (#37)

* Basic structure for eval correlation script

* calculate correlation

---------

Co-authored-by: astrophie <sophie.w.feng@gmail.com>

examples/evaluate_annotation.py

@@ -0,0 +1,67 @@
+import pandas as pd
+import rich
+import typer
+from scipy.stats import pearsonr
+from sotopia.database import EpisodeLog
+from sotopia.database.serialization import get_rewards_from_episode
+from typer import Typer
+
+app = Typer()
+
+
+def get_rewards_from_episodes(
+    episode_pks: list[str], categories: list[str]
+) -> dict[str, list[float]]:
+    rewards: dict[str, list[float]] = {category: [] for category in categories}
+    for episode_pk in episode_pks:
+        episode = EpisodeLog.get(episode_pk)
+        reward = get_rewards_from_episode(episode)
+        reward_agent_2 = reward[1][1]
+        for category in categories:
+            rewards[category].append(float(reward_agent_2[category]))
+    return rewards
+
+
+@app.command()
+def calculate_pearson_correlation_and_accuracy(csv_path: str) -> None:
+    df = pd.read_csv(csv_path)
+
+    categories = [
+        "targeted_safety_risks",
+        "system_and_operational_risks",
+        "content_safety_risks",
+        "societal_risks",
+        "legal_and_rights_related_risks",
+        "efficiency",
+        "goal",
+    ]
+    rewards = get_rewards_from_episodes(df["episode_pk"].to_list(), categories)
+    correlations: dict[str, tuple[float, float]] = {}
+    for category in categories:
+        if category in df.columns:
+            result = pearsonr(df[category].astype(float), rewards[category])
+            correlations[category] = (result.statistic, result.pvalue)
+        else:
+            correlations[category] = (0.0, 0.0)
+
+    for category, (corr, p_value) in correlations.items():
+        print(f"{category}: Pearson correlation = {corr}, p-value = {p_value}")
+
+    # Calculate accuracy
+    accuracy = {}
+    for category in categories:
+        if category in df.columns:
+            accuracy[category] = (
+                (pd.Series(df[category]).astype(float) != 0)
+                .eq(pd.Series(rewards[category]).astype(float) != 0)
+                .mean()
+            )
+        else:
+            accuracy[category] = 0.0
+
+    for category, acc in accuracy.items():
+        print(f"{category}: Accuracy = {acc}")
+
+
+if __name__ == "__main__":
+    app()