Update app.py

app.py

@@ -2,92 +2,128 @@ import json
 import numpy as np
 from fastapi import FastAPI
 from pydantic import BaseModel
-from huggingface_hub import hf_hub_download
 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")
+
+# ============================================================
+# Model Registry — maps category → HF repo + file
+# ============================================================
 
-# ===============================
-# REPO + FOLDERS (REAL STRUCTURE)
-# ===============================
-REPO_ID = "ClergeF/MVT-models"
-EMBEDDER_FOLDER = "universal_embedder"
-MODEL_FOLDER = "models"
+HF_USER = "ClergeF"
+
+MODEL_REPOS = {
+    "value":       "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"
+}
 
-# ===============================
-# JSON FILES FOR YOUR 12 MODELS
-# ===============================
 MODEL_FILES = {
-    "value_impact": "value_impact.json",
-    "impact": "impact.json",
-    "family": "family_level.json",
-    "community": "community_level.json",
-    "education": "education_level.json",
-    "health": "health_level.json",
+    "value":       "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 EMBEDDER (NO RENAMING)
-# ===============================
-embedder = SentenceTransformer("ClergeF/MVT-embedder")
+# ============================================================
+# Load all category models into memory
+# ============================================================
+
+loaded_models = {}
+
+def load_single_model(category: str):
+    """Download & load one model's JSON coefficients."""
+    repo = MODEL_REPOS[category]
+    file = MODEL_FILES[category]
 
-# ===============================
-# LOAD ALL MODELS
-# ===============================
-def load_json_model(filename):
     path = hf_hub_download(
-        repo_id=REPO_ID,
-        filename=f"{MODEL_FOLDER}/{filename}"
+        repo_id=f"{HF_USER}/{repo}",
+        filename=file
     )
+
     with open(path, "r") as f:
-        return json.load(f)
+        data = json.load(f)
 
-models = {key: load_json_model(file) for key, file in MODEL_FILES.items()}
+    model = {
+        "weights": np.array(data["weights"]),
+        "bias": float(data["bias"])
+    }
+    return model
 
-# ===============================
-# FASTAPI SETUP
-# ===============================
-app = FastAPI(title="MVT Category + Impact API")
 
-class Input(BaseModel):
+print("Loading all 12 models...")
+for cat in MODEL_REPOS:
+    loaded_models[cat] = load_single_model(cat)
+print("All models loaded successfully.")
+
+
+# ============================================================
+# Input schema
+# ============================================================
+
+class InputText(BaseModel):
     text: str
 
-# ===============================
-# HELPERS
-# ===============================
-def embed(text):
-    return embedder.encode([text])[0]
-
-def linear_predict(model, vec):
-    coef = np.array(model["coef"])
-    intercept = np.array(model["intercept"])
-    if coef.ndim == 2:
-        return coef @ vec + intercept
-    return float(np.dot(coef, vec) + intercept)
-
-# ===============================
-# API ROUTE
-# ===============================
+
+# ============================================================
+# Predict function per model
+# ============================================================
+
+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
+
+
+# ============================================================
+# API Routes
+# ============================================================
+
+@app.get("/")
+def root():
+    return {"message": "MVT Category API is running."}
+
+
 @app.post("/predict")
-def predict_text(data: Input):
-    text = data.text
-    vec = embed(text)
+def predict(payload: InputText):
+    text = payload.text
 
-    output = {}
+    results = {}
 
-    # Value + Impact (2-output regression)
-    value, imp = linear_predict(models["value_impact"], vec)
-    output["estimated_value"] = float(value)
-    output["impact_level"] = float(imp)
+    for category, model in loaded_models.items():
+        results[category] = predict_single(text, model)
 
-    # All 10 category levels + single impact model
-    for key in MODEL_FILES.keys():
-        if key == "value_impact":
-            continue
-        output[key] = float(linear_predict(models[key], vec))
+    # Also return categories >= 0.85
+    high_cats = [c for c, s in results.items() if s >= 0.85]
 
-    return {"input": text, "predictions": output}
+    return {
+        "input": text,
+        "scores": results,
+        "high_confidence_categories": high_cats
+    }