Upload assets/config/best-performing-model_config.json with huggingface_hub

assets/config/best-performing-model_config.json

@@ -0,0 +1,38 @@
+{
+    "model_name": "OscarNominationPredictor",
+    "encoder": "intfloat/e5-base-v2",
+    "encoder_usage": {
+        "prefix": "query:",
+        "normalize": true,
+        "dtype": "float16 (GPU) -> float32 (disk)"
+    },
+    "input_fields": [
+        "script_clean",
+        "summary",
+        "title"
+    ],
+    "chunking": {
+        "max_words": 400,
+        "overlap": 80
+    },
+    "pooling": "mean + max concatenation + L2",
+    "classifier": {
+        "type": "LogisticRegression",
+        "class_weight": "balanced",
+        "C": 1.0,
+        "max_iter": 5000
+    },
+    "threshold_selection": {
+        "metric": "F1 (positive class)",
+        "best_threshold_val": 0.52
+    },
+    "test_metrics": {
+        "accuracy": 0.759,
+        "roc_auc": 0.79,
+        "pr_auc": 0.455,
+        "f1_positive": 0.485,
+        "f1_negative": 0.843,
+        "macro_f1": 0.664
+    },
+    "notes": "Best performing logistic regression model using concatenated script+summary+title embeddings (E5-base-v2) with mean+max pooling. Threshold chosen to maximize F1 on validation set."
+}