Upload app.py with huggingface_hub

app.py

@@ -163,7 +163,7 @@ def nse_kategori(v):
 
 
 # ── Tab 1: Prediksi ────────────────────────────────────────────────────────────
-def jalankan_prediksi(station_label: str, n_history_days: int):
+def jalankan_prediksi(station_label: str, n_history_days: int, periode: str):
     try:
         station_id = [k for k, v in STATION_LABELS.items() if v == station_label][0]
         df = _load_data()
@@ -172,10 +172,34 @@ def jalankan_prediksi(station_label: str, n_history_days: int):
         from pytorch_forecasting import TimeSeriesDataSet
 
         df_station = df[df["station_id"] == station_id].reset_index(drop=True)
-        last_idx   = df_station["time_idx"].max()
-        start_idx  = last_idx - MAX_ENCODER_LENGTH - MAX_PREDICTION_LENGTH + 1
-        df_window  = df[
-            (df["station_id"] == station_id) & (df["time_idx"] >= start_idx)
+        n_steps      = df_station["time_idx"].max() + 1
+        train_cutoff = int(n_steps * 0.75)
+
+        # Pilih window berdasarkan periode
+        # Gunakan time_idx absolut dari df asli (bukan reset index)
+        df_station_full = df[df["station_id"] == station_id]
+        abs_max_idx = df_station_full["time_idx"].max()
+        abs_min_val_idx = df_station_full[
+            df_station_full["time_idx"] > train_cutoff
+        ]["time_idx"].min()
+
+        PERIODE_OFFSETS = {
+            "Awal Validasi (Okt 2022)":  abs_min_val_idx + MAX_ENCODER_LENGTH,
+            "Tengah Validasi (Apr 2023)": abs_min_val_idx + int((abs_max_idx - abs_min_val_idx) * 0.4),
+            "Akhir Validasi (Des 2023)":  abs_max_idx - MAX_PREDICTION_LENGTH,
+        }
+        anchor_idx = PERIODE_OFFSETS.get(periode, PERIODE_OFFSETS["Tengah Validasi (Apr 2023)"])
+        # anchor_idx = titik akhir encoder (mulai prediction dari sini)
+        anchor_idx = max(anchor_idx, abs_min_val_idx + MAX_ENCODER_LENGTH)
+        anchor_idx = min(anchor_idx, abs_max_idx - MAX_PREDICTION_LENGTH)
+
+        start_idx = anchor_idx - MAX_ENCODER_LENGTH
+        end_idx   = anchor_idx + MAX_PREDICTION_LENGTH
+
+        df_window = df[
+            (df["station_id"] == station_id) &
+            (df["time_idx"] >= start_idx) &
+            (df["time_idx"] <  end_idx)
         ].copy()
 
         pred_dataset = TimeSeriesDataSet.from_dataset(
@@ -596,6 +620,15 @@ with gr.Blocks(
                         value=STATION_LABELS["sutami"],
                         label="📍 Pilih Stasiun / Bendungan",
                     )
+                    periode_dd = gr.Dropdown(
+                        choices=[
+                            "Awal Validasi (Okt 2022)",
+                            "Tengah Validasi (Apr 2023)",
+                            "Akhir Validasi (Des 2023)",
+                        ],
+                        value="Tengah Validasi (Apr 2023)",
+                        label="📆 Pilih Periode Prediksi",
+                    )
                     history_sl = gr.Slider(
                         minimum=1, maximum=7, value=3, step=1,
                         label="📅 Tampilkan histori (hari)",
@@ -609,7 +642,7 @@ with gr.Blocks(
 
             run_btn.click(
                 fn=jalankan_prediksi,
-                inputs=[station_dd, history_sl],
+                inputs=[station_dd, history_sl, periode_dd],
                 outputs=[plot_out, metrik_out],
             )