Update app.py

app.py

@@ -2,128 +2,146 @@ import json
 import numpy as np
 from fastapi import FastAPI
 from pydantic import BaseModel
-from sentence_transformers import SentenceTransformer
 from huggingface_hub import hf_hub_download
-import os
-
-app = FastAPI()
-
-# ============================================================
-# Load Embedder  (NO MORE CUSTOM REPO — USE BASE MODEL)
-# ============================================================
-
-print("Loading embedder: all-MiniLM-L6-v2 ...")
-embedder = SentenceTransformer("all-MiniLM-L6-v2")
+from sentence_transformers import SentenceTransformer
 
 # ============================================================
-# Model Registry — maps category → HF repo + file
+# CONFIG
 # ============================================================
 
 HF_USER = "ClergeF"
 
+# Your 12 individual HF repos
 MODEL_REPOS = {
-    "value":       "value-impact-model",
-    "impact":      "impact-model",
-    "family":      "family-model",
-    "community":   "community-model",
-    "education":   "education-model",
-    "health":      "health-model",
+    "value_impact": "value-impact-model",
+    "impact": "impact-model",
+    "family": "family-model",
+    "community": "community-model",
+    "education": "education-model",
+    "health": "health-model",
     "environment": "environment-model",
-    "business":    "business-model",
-    "finance":     "finance-model",
-    "history":     "history-model",
-    "spirituality":"spirituality-model",
-    "innovation":  "innovation-model"
+    "business": "business-model",
+    "finance": "finance-model",
+    "history": "history-model",
+    "spirituality": "spirituality-model",
+    "innovation": "innovation-model",
 }
 
+# Model file names inside each repo
 MODEL_FILES = {
-    "value":       "value_impact.json",
-    "impact":      "impact.json",
-    "family":      "family_level.json",
-    "community":   "community_level.json",
-    "education":   "education_level.json",
-    "health":      "health_level.json",
+    "value_impact": "value_impact.json",
+    "impact": "impact.json",
+    "family": "family_level.json",
+    "community": "community_level.json",
+    "education": "education_level.json",
+    "health": "health_level.json",
     "environment": "environment_level.json",
-    "business":    "business_level.json",
-    "finance":     "finance_level.json",
-    "history":     "history_level.json",
-    "spirituality":"spirituality_level.json",
-    "innovation":  "innovation_level.json"
+    "business": "business_level.json",
+    "finance": "finance_level.json",
+    "history": "history_level.json",
+    "spirituality": "spirituality_level.json",
+    "innovation": "innovation_level.json",
 }
 
 # ============================================================
-# Load all category models into memory
+# LOAD EMBEDDER
 # ============================================================
 
-loaded_models = {}
+# Keep it light + consistent for Spaces
+print("Loading embedder: all-MiniLM-L6-v2 …")
+embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 
-def load_single_model(category: str):
-    """Download & load one model's JSON coefficients."""
-    repo = MODEL_REPOS[category]
-    file = MODEL_FILES[category]
+
+# ============================================================
+# MODEL LOADING HELPERS
+# ============================================================
+
+def load_model_from_hf(repo_name: str, filename: str):
+    """Downloads a model.json file from HuggingFace and returns Python dict."""
+    print(f"↳ Loading {filename} from {repo_name} …")
 
     path = hf_hub_download(
-        repo_id=f"{HF_USER}/{repo}",
-        filename=file
+        repo_id=f"{HF_USER}/{repo_name}",
+        filename=filename
     )
 
     with open(path, "r") as f:
-        data = json.load(f)
+        return json.load(f)
+
+
+def embed(text: str):
+    """Returns a 384-dim sentence embedding."""
+    return embedder.encode([text])[0]
 
-    model = {
-        "weights": np.array(data["weights"]),
-        "bias": float(data["bias"])
-    }
-    return model
 
+def linear_predict(model_json, vec):
+    """Linear model forward pass using coef + intercept."""
+    coef = np.array(model_json["coef"])
+    intercept = np.array(model_json["intercept"])
 
-print("Loading all 12 models...")
-for cat in MODEL_REPOS:
-    loaded_models[cat] = load_single_model(cat)
-print("All models loaded successfully.")
+    if coef.ndim == 2:      # Multi-output
+        return coef @ vec + intercept
+    else:
+        return float(np.dot(coef, vec) + intercept)
 
 
 # ============================================================
-# Input schema
+# LOAD ALL 12 MODELS AT STARTUP
 # ============================================================
 
-class InputText(BaseModel):
-    text: str
+print("Loading all 12 models…")
 
+loaded_models = {}
 
-# ============================================================
-# Predict function per model
-# ============================================================
+for key in MODEL_REPOS:
+    repo = MODEL_REPOS[key]
+    file = MODEL_FILES[key]
+
+    model_json = load_model_from_hf(repo, file)
+    loaded_models[key] = model_json
 
-def predict_single(text: str, model_dict):
-    embedding = embedder.encode([text])[0]          # vector
-    score = float(np.dot(embedding, model_dict["weights"]) + model_dict["bias"])
-    return max(0.0, min(1.0, score))                # clamp 0–1
+print("✔ All models loaded successfully.")
 
 
 # ============================================================
-# API Routes
+# FASTAPI APP
 # ============================================================
 
+app = FastAPI(title="MVT Community Value API (Production)")
+
+
+class InputText(BaseModel):
+    text: str
+
+
 @app.get("/")
 def root():
-    return {"message": "MVT Category API is running."}
+    return {"status": "ok", "message": "MVT Category & Value API running!"}
 
 
+# ============================================================
+# PREDICT ROUTE
+# ============================================================
+
 @app.post("/predict")
 def predict(payload: InputText):
     text = payload.text
+    vec = embed(text)
 
-    results = {}
+    out = {}
 
-    for category, model in loaded_models.items():
-        results[category] = predict_single(text, model)
+    # MULTI-OUTPUT: Value + Impact
+    value_pred, impact_pred = linear_predict(loaded_models["value_impact"], vec)
+    out["estimated_value"] = float(value_pred)
+    out["impact_level"] = float(impact_pred)
 
-    # Also return categories >= 0.85
-    high_cats = [c for c, s in results.items() if s >= 0.85]
+    # ALL OTHER SINGLE-OUTPUT MODELS
+    for key in MODEL_REPOS:
+        if key == "value_impact":
+            continue  # skip, already handled
+        out[key] = float(linear_predict(loaded_models[key], vec))
 
     return {
         "input": text,
-        "scores": results,
-        "high_confidence_categories": high_cats
+        "predictions": out
     }