Update app.py

app.py

@@ -1,91 +1,87 @@
+import json
+import numpy as np
 from fastapi import FastAPI
 from pydantic import BaseModel
 from huggingface_hub import hf_hub_download
-import json
-
-app = FastAPI(
-    title="Impact Rating API",
-    description="Rates an action using your impact.json scoring rules",
-    version="1.0"
-)
-
-# ---------------------------------------------------------
-# Load the JSON model from HuggingFace Hub
-# ---------------------------------------------------------
+from sentence_transformers import SentenceTransformer
 
-REPO_ID = "ClergeF/impact-model"
-FILENAME = "impact.json"
+# ============================================================
+# CONFIG
+# ============================================================
 
-print("Downloading impact.json from Hugging Face…")
+HF_USER = "ClergeF"
+IMPACT_REPO = "impact-model"
+IMPACT_FILE = "impact.json"
 
-json_path = hf_hub_download(
-    repo_id=REPO_ID,
-    filename=FILENAME,
-    repo_type="model"
-)
+# ============================================================
+# EMBEDDER
+# ============================================================
 
-with open(json_path, "r") as f:
-    impact_model = json.load(f)
+print("Loading embedder: all-MiniLM-L6-v2 …")
+embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 
-print("Impact model loaded successfully!")
+def embed(text: str):
+    return embedder.encode([text])[0]
 
+# ============================================================
+# LOAD IMPACT MODEL
+# ============================================================
 
-# ---------------------------------------------------------
-# Request Body
-# ---------------------------------------------------------
-
-class Input(BaseModel):
-    text: str
+def load_model(repo, filename):
+    print(f"Loading: {repo}/{filename}")
+    path = hf_hub_download(
+        repo_id=f"{HF_USER}/{repo}",
+        filename=filename,
+        revision="main",
+        force_download=True
+    )
+    with open(path, "r") as f:
+        return json.load(f)
 
+print("Loading impact model…")
+impact_model = load_model(IMPACT_REPO, IMPACT_FILE)
+print("✔ Impact model loaded!")
 
-# ---------------------------------------------------------
-# Rating Function
-# ---------------------------------------------------------
+# ============================================================
+# LINEAR PREDICT
+# ============================================================
 
-def compute_impact(text: str):
-    """
-    Super simple rule engine:
-    - Look through `impact_model`
-    - If text contains a keyword → return score
-    - If none match → return neutral
-    """
+def linear_predict(model_json, vec):
+    coef = np.array(model_json["coef"])
+    intercept = np.array(model_json["intercept"])
+    return float(np.dot(coef, vec) + intercept)
 
-    text_lower = text.lower()
+# ============================================================
+# API
+# ============================================================
 
-    for rule in impact_model.get("rules", []):
-        keywords = rule.get("keywords", [])
-        score = rule.get("score", 0)
+app = FastAPI(title="Impact Rating API")
 
-        for kw in keywords:
-            if kw.lower() in text_lower:
-                return {
-                    "impact_score": score,
-                    "matched_keyword": kw
-                }
-
-    # default fallback
-    return {
-        "impact_score": impact_model.get("default_score", 0),
-        "matched_keyword": None
-    }
-
-
-# ---------------------------------------------------------
-# Endpoints
-# ---------------------------------------------------------
+class InputText(BaseModel):
+    text: str
 
 @app.get("/")
 def home():
-    return {
-        "message": "Impact Model API Running",
-        "model": REPO_ID,
-        "status": "ok"
-    }
+    return {"status": "ok", "message": "Impact API running"}
 
 @app.post("/rate")
-def rate(item: Input):
-    result = compute_impact(item.text)
+def rate(payload: InputText):
+    text = payload.text
+    vec = embed(text)
+
+    try:
+        score = linear_predict(impact_model, vec)
+        result = {
+            "impact_score": score,
+            "matched_keyword": None
+        }
+    except Exception as e:
+        result = {
+            "impact_score": impact_model.get("default_score", 0),
+            "matched_keyword": None
+        }
+
     return {
-        "input": item.text,
+        "input": text,
         "result": result
     }