Update app.py

app.py

@@ -2,37 +2,19 @@ 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
-
-# ============================================================
-# CONFIG — UPDATE ONLY IF YOU CHANGE REPO NAMES
-# ============================================================
-
-# Your username
-HF_USER = "ClergeF"
-
-# Your embedder repo
-EMBEDDER_REPO = "MVT-embedder"
+from sentence_transformers import SentenceTransformer
 
-# Your 12 model repos
-MODEL_REPOS = {
-    "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",
-}
+# ===============================
+# REPO + FOLDERS (REAL STRUCTURE)
+# ===============================
+REPO_ID = "ClergeF/MVT-models"
+EMBEDDER_FOLDER = "universal_embedder"
+MODEL_FOLDER = "models"
 
-# Files inside each repo
+# ===============================
+# JSON FILES FOR YOUR 12 MODELS
+# ===============================
 MODEL_FILES = {
     "value_impact": "value_impact.json",
     "impact": "impact.json",
@@ -48,101 +30,67 @@ MODEL_FILES = {
     "innovation": "innovation_level.json",
 }
 
-# ============================================================
-# LOAD EMBEDDER (automatically works in Spaces)
-# ============================================================
-
-print("Loading SentenceTransformer embedder...")
-embedder = SentenceTransformer(f"{HF_USER}/{EMBEDDER_REPO}")
-
-# ============================================================
-# MODEL LOADING HELPERS
-# ============================================================
-
-def load_json_model(repo_name: str, filename: str):
-    """Download and load model JSON from HuggingFace Hub."""
-    print(f"Loading {repo_name}/{filename} ...")
-
+# ===============================
+# LOAD EMBEDDER (NO RENAMING)
+# ===============================
+embedder = SentenceTransformer(
+    REPO_ID,
+    subfolder=EMBEDDER_FOLDER
+)
+
+# ===============================
+# LOAD ALL MODELS
+# ===============================
+def load_json_model(filename):
     path = hf_hub_download(
-        repo_id=f"{HF_USER}/{repo_name}",
-        filename=filename
+        repo_id=REPO_ID,
+        filename=f"{MODEL_FOLDER}/{filename}"
     )
-
     with open(path, "r") as f:
         return json.load(f)
 
-# ============================================================
-# LOAD ALL 12 MODELS ONCE
-# ============================================================
+models = {key: load_json_model(file) for key, file in MODEL_FILES.items()}
 
-print("Loading linear regression weight files...")
-
-models = {}
-
-for key, repo in MODEL_REPOS.items():
-    models[key] = load_json_model(repo, MODEL_FILES[key])
-
-print("All models successfully loaded!")
-
-# ============================================================
+# ===============================
 # FASTAPI SETUP
-# ============================================================
-
-app = FastAPI(title="MVT Category Scoring API")
+# ===============================
+app = FastAPI(title="MVT Category + Impact API")
 
-class InputText(BaseModel):
+class Input(BaseModel):
     text: str
 
-# ============================================================
-# PREDICTION HELPERS
-# ============================================================
-
-def embed(text: str):
-    """Convert text → embedding."""
+# ===============================
+# HELPERS
+# ===============================
+def embed(text):
     return embedder.encode([text])[0]
 
-def linear_predict(model_json, vec):
-    """Apply manual linear regression: dot(coef, x) + intercept."""
-    coef = np.array(model_json["coef"])
-    intercept = model_json["intercept"]
-
-    if isinstance(intercept, list):
-        intercept = np.array(intercept)
-
-    pred = np.dot(coef, vec) + intercept
-    return float(pred)
-
-# ============================================================
-# MAIN API ROUTE
-# ============================================================
+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
+# ===============================
 @app.post("/predict")
-def predict(payload: InputText):
-    text = payload.text
+def predict_text(data: Input):
+    text = data.text
     vec = embed(text)
 
-    result = {}
-
-    # --- Value + Impact model (2 outputs) ---
-    value_impact_model = models["value_impact"]
-    coef = np.array(value_impact_model["coef"])
-    intercept = np.array(value_impact_model["intercept"])
-
-    value_pred, impact_pred = coef @ vec + intercept
-    result["estimated_value"] = float(value_pred)
-    result["impact_level"] = float(impact_pred)
+    output = {}
 
-    # --- 10 category models ---
-    for name in [
-        "family", "community", "education", "health", "environment",
-        "business", "finance", "history", "spirituality", "innovation"
-    ]:
-        result[f"{name}_score"] = linear_predict(models[name], vec)
+    # Value + Impact (2-output regression)
+    value, imp = linear_predict(models["value_impact"], vec)
+    output["estimated_value"] = float(value)
+    output["impact_level"] = float(imp)
 
-    # --- Standalone impact model ---
-    result["impact_model_score"] = linear_predict(models["impact"], vec)
+    # 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))
 
-    return {
-        "input": text,
-        "predictions": result
-    }
+    return {"input": text, "predictions": output}