Update Gradio app with multiple files

model_handler.py

@@ -1,6 +1,15 @@
 import numpy as np
 import torch
-from chronos import BaseChronosPipeline 
+import warnings
+
+# Make chronos import optional
+try:
+    from chronos import BaseChronosPipeline
+    CHRONOS_AVAILABLE = True
+except ImportError:
+    warnings.warn("Chronos-forecasting not available. Using fallback predictions.")
+    CHRONOS_AVAILABLE = False
+    BaseChronosPipeline = None
 
 class ModelHandler:
     def __init__(self):
@@ -11,6 +20,10 @@ class ModelHandler:
     
     def load_model(self):
         """Load Chronos-2 model using the official BaseChronosPipeline"""
+        if not CHRONOS_AVAILABLE:
+            print("Chronos-forecasting not installed. Using fallback prediction method.")
+            return
+            
         try:
             print(f"Loading {self.model_name} on {self.device}...")
             
@@ -27,30 +40,18 @@ class ModelHandler:
     
     def predict(self, data, horizon=10):
         """Generate predictions using Chronos-2 or fallback."""
+        if not CHRONOS_AVAILABLE or self.pipeline is None:
+            # Fallback to simple trend-based prediction
+            return self._fallback_predict(data, horizon)
+            
         try:
             if data is None or not isinstance(data, dict) or 'original' not in data or len(data['original']) < 20:
-                return np.array([0] * horizon)
-            
-            if self.pipeline is None:
-                # --- Fallback Logic ---
-                values = data['original']
-                recent_trend = np.polyfit(range(len(values[-20:])), values[-20:], 1)[0]
-                
-                predictions = []
-                last_value = values[-1]
-                
-                for i in range(horizon):
-                    next_value = last_value + recent_trend * (i + 1)
-                    noise = np.random.normal(0, data.get('std', 1.0) * 0.1)
-                    predictions.append(next_value + noise)
-                
-                return np.array(predictions)
+                return self._fallback_predict(data, horizon)
             
             # --- Chronos-2 Inference ---
             predictions_samples = self.pipeline.predict(
                 data['original'],
                 prediction_length=horizon,
-                # FIX UTAMA: Menghapus 'n_samples'
             )
             
             # Mengambil nilai rata-rata (mean) dari semua sampel atau single trajectory
@@ -64,5 +65,32 @@ class ModelHandler:
             return mean_predictions
             
         except Exception as e:
-            print(f"Prediction error: {e}")
-            return np.array([0] * horizon)
+            print(f"Prediction error with Chronos: {e}. Using fallback.")
+            return self._fallback_predict(data, horizon)
+    
+    def _fallback_predict(self, data, horizon=10):
+        """Fallback prediction method when Chronos is unavailable"""
+        try:
+            if data is None or not isinstance(data, dict) or 'original' not in data:
+                # Return zero predictions if no data
+                return np.zeros(horizon)
+            
+            values = data['original']
+            if len(values) < 5:
+                return np.zeros(horizon)
+            
+            # Simple trend extrapolation
+            recent_trend = np.polyfit(range(len(values[-20:])), values[-20:], 1)[0]
+            predictions = []
+            last_value = values[-1]
+            
+            for i in range(horizon):
+                next_value = last_value + recent_trend * (i + 1)
+                noise = np.random.normal(0, data.get('std', 1.0) * 0.1)
+                predictions.append(next_value + noise)
+            
+            return np.array(predictions)
+            
+        except Exception as e:
+            print(f"Fallback prediction error: {e}")
+            return np.zeros(horizon)

requirements.txt

@@ -1,4 +1,3 @@
-=== requirements.txt ===
 pandas
 plotly
 numpy
@@ -18,4 +17,6 @@ tokenizers
 yfinance
 scipy
 joblib
-chronos-forecasting
+chronos-forecasting
+safetensors
+huggingface-hub