Upload 2 files

app.py

@@ -0,0 +1,76 @@
+from fastapi import FastAPI, HTTPException
+from pydantic import BaseModel, Field
+import numpy as np
+import tensorflow as tf
+import joblib
+import os
+from typing import List, Dict
+
+app = FastAPI(title="Multimodal T5 Classification API")
+
+# Define paths (use defaults set in training script)
+MODEL_PATH = "./output/T5_best_multimodel.keras"
+SCALER_PATH = "./output/scaler.pkl"
+
+# Global variables for model and scaler
+model = None
+scaler = None
+
+@app.on_event("startup")
+def load_artifacts():
+    global model, scaler
+    if not os.path.exists(MODEL_PATH):
+        raise RuntimeError(f"Model file not found at {MODEL_PATH}. Please run training first.")
+    if not os.path.exists(SCALER_PATH):
+        raise RuntimeError(f"Scaler file not found at {SCALER_PATH}. Please run training first.")
+    
+    model = tf.keras.models.load_model(MODEL_PATH)
+    scaler = joblib.load(SCALER_PATH)
+    print("Model and scaler loaded successfully.")
+
+class FeaturesInput(BaseModel):
+    # Total 17 features as used in standard scaler + normalization
+    # Column indices based on notebook: X.iloc[:,3:]
+    # Original columns in Excel were 17 (3:20 range)
+    features: List[float] = Field(..., min_items=17, max_items=17, 
+                                 description="List of 17 numeric features in original scale")
+
+@app.post("/predict")
+async def predict(input_data: FeaturesInput):
+    if model is None or scaler is None:
+        raise HTTPException(status_code=500, detail="Model artifacts not loaded.")
+    
+    try:
+        # 1. Scaling
+        raw_features = np.array(input_data.features).reshape(1, -1)
+        scaled_features = scaler.transform(raw_features)
+        
+        # 2. Modality Splitting
+        # Visual features: columns 1-6 + column 0 (of relative subset)
+        # In our case, the 17 features correspond to X.iloc[:,3:]
+        # visual_features = np.concatenate((X.iloc[:,1:7].to_numpy(), X.iloc[:,0].to_numpy().reshape(-1,1)), axis=1)
+        # speech_features = np.concatenate((X.iloc[:,7:11].to_numpy(), X.iloc[:,0].to_numpy().reshape(-1,1)), axis=1)
+        # pupl_features =  np.concatenate((X.iloc[:,11:17].to_numpy(), X.iloc[:,0].to_numpy().reshape(-1,1)), axis=1)
+        
+        f = scaled_features[0]
+        visual = np.concatenate([f[1:7], [f[0]]]).reshape(1, -1)
+        speech = np.concatenate([f[7:11], [f[0]]]).reshape(1, -1)
+        pupl = np.concatenate([f[11:17], [f[0]]]).reshape(1, -1)
+        
+        # 3. Inference
+        prediction = model.predict([visual, speech, pupl], verbose=0)
+        predicted_class = int(np.argmax(prediction, axis=1)[0])
+        probabilities = prediction[0].tolist()
+        
+        return {
+            "predicted_class": predicted_class,
+            "confidence": probabilities[predicted_class],
+            "probabilities": {str(i): prob for i, prob in enumerate(probabilities)}
+        }
+        
+    except Exception as e:
+        raise HTTPException(status_code=400, detail=str(e))
+
+@app.get("/health")
+def health():
+    return {"status": "ok", "model_loaded": model is not None}

requirements.txt

@@ -0,0 +1,10 @@
+numpy
+pandas
+scikit-learn
+joblib
+keras
+matplotlib
+seaborn
+tensorflow
+uvicorn[standard]
+fastapi