Atualizar uso de NaN em predição

predict.py

@@ -9,6 +9,7 @@ import tensorflow as tf
 import matplotlib.pyplot as plt
 import base64
 from io import BytesIO
+import math
 
 warnings.filterwarnings('ignore')
 plt.style.use('seaborn-v0_8-darkgrid')
@@ -101,6 +102,7 @@ class DenguePredictor:
         for i in range(weeks_to_predict):
             dynamic_input = np.array([dynamic_sequence_scaled], dtype=np.float32)
             pred_scaled = self.model.predict([dynamic_input, static_input], verbose=0)[0][0]
+
             pred_real = self.inverse_transform_cases(scaler_dyn, np.array([pred_scaled]))[0]
 
             future_date = last_date + timedelta(weeks=i + 1)
@@ -109,25 +111,19 @@ class DenguePredictor:
                 "predicted_cases": max(0, round(pred_real))
             })
 
-            # Seleção de clima futuro
-            last_year_date = future_date - timedelta(weeks=52)
-            target_week = last_year_date.isocalendar()[1]
-            target_year = last_year_date.year
-            if (target_year, target_week) in climate_lookup:
-                future_climate = climate_lookup[(target_year, target_week)]
+            future_year, future_week = future_date.year, future_date.isocalendar()[1]
+            if (future_year, future_week) in climate_lookup:
+                future_climate = climate_lookup[(future_year, future_week)]
             else:
-                # Fallback: média das últimas 4 semanas
-                future_climate = df_mun[dynamic_features].tail(4).mean()
-
-            # Atualizar sequência dinâmica
-            new_row = {
-                "numero_casos": pred_scaled,
-                "T2M": future_climate["T2M"], "T2M_MAX": future_climate["T2M_MAX"],
-                "T2M_MIN": future_climate["T2M_MIN"], "PRECTOTCORR": future_climate["PRECTOTCORR"],
-                "RH2M": future_climate["RH2M"], "ALLSKY_SFC_SW_DWN": future_climate["ALLSKY_SFC_SW_DWN"]
-            }
-            new_row_df = pd.DataFrame([new_row])[dynamic_features]
-            new_row_scaled = scaler_dyn.transform(new_row_df)[0]
+                future_climate = df_mun[dynamic_features[1:]].tail(4).mean()  # só clima, sem numero_casos
+
+            new_row = np.zeros(len(dynamic_features), dtype=np.float32)
+            new_row[0] = pred_scaled  # numero_casos previsto
+            for j, feat in enumerate(dynamic_features[1:], start=1):
+                new_row[j] = future_climate[feat]
+
+            # 6) Escala a linha futura e atualiza sequência
+            new_row_scaled = scaler_dyn.transform(new_row.reshape(1, -1))[0]
             dynamic_sequence_scaled = np.vstack([dynamic_sequence_scaled[1:], new_row_scaled])
 
         # Histórico das últimas 52 semanas