Patch sweep notebook to avoid sklearn numpy import error

stanceeval2026_smoke_sweep_ensemble.py

@@ -6,13 +6,12 @@ import os, json, random, subprocess, sys
 from pathlib import Path
 
 # Install dependencies when running in Colab or a fresh environment.
-subprocess.run([sys.executable, "-m", "pip", "install", "-q", "-U", "transformers>=4.57.0", "datasets", "evaluate", "accelerate", "scikit-learn", "pandas", "numpy", "huggingface_hub"], check=True)
+subprocess.run([sys.executable, "-m", "pip", "install", "-q", "-U", "transformers>=4.57.0", "datasets", "evaluate", "accelerate", "pandas", "numpy<2.0", "huggingface_hub"], check=True)
 
 import numpy as np
 import pandas as pd
 import torch
 from datasets import Dataset
-from sklearn.metrics import f1_score, accuracy_score
 from transformers import AutoTokenizer, AutoConfig, AutoModelForSequenceClassification, DataCollatorWithPadding, TrainingArguments, Trainer, set_seed
 
 SEED = 42
@@ -92,11 +91,22 @@ train_ds_full = to_hf_dataset(train_df)
 dev_ds_full = to_hf_dataset(dev_df)
 
 # Metrics: Favg2 is macro-F1 over Favor and Against.
+def _f1_for_label(y_true, y_pred, label):
+    y_true = np.asarray(y_true)
+    y_pred = np.asarray(y_pred)
+    tp = int(((y_true == label) & (y_pred == label)).sum())
+    fp = int(((y_true != label) & (y_pred == label)).sum())
+    fn = int(((y_true == label) & (y_pred != label)).sum())
+    denom = 2 * tp + fp + fn
+    return 0.0 if denom == 0 else (2 * tp / denom)
+
 def stance_metrics(y_true, y_pred):
-    per = f1_score(y_true, y_pred, labels=[0, 1, 2], average=None, zero_division=0)
+    y_true = np.asarray(y_true)
+    y_pred = np.asarray(y_pred)
+    per = np.array([_f1_for_label(y_true, y_pred, 0), _f1_for_label(y_true, y_pred, 1), _f1_for_label(y_true, y_pred, 2)], dtype=float)
     return {
-        'accuracy': float(accuracy_score(y_true, y_pred)),
-        'macro_f1': float(f1_score(y_true, y_pred, average='macro', zero_division=0)),
+        'accuracy': float((y_true == y_pred).mean()),
+        'macro_f1': float(np.mean(per)),
         'favg2': float((per[0] + per[1]) / 2.0),
         'favg3': float(np.mean(per)),
         'favor_f1': float(per[0]),