refactor: reorganize project structure by moving core modules and update import paths in API server

core/ndvi_data_loader.py

@@ -0,0 +1,93 @@
+import os
+import torch
+import numpy as np
+import xarray as xr
+from torch.utils.data import Dataset
+import glob
+
+class NDVITimeSeriesDataset(Dataset):
+    def __init__(self, sequence_length=3, spatial=False):
+        """
+        Đọc dữ liệu S2 từ cache, tính NDVI và tạo Time-Series.
+        spatial=False -> Output 1D cho LSTM/ARIMA
+        spatial=True -> Output 2D cho ConvLSTM
+        """
+        self.sequence_length = sequence_length
+        self.spatial = spatial
+        self.data_seqs = []
+        self.targets = []
+        
+        # Load from cache
+        cache_files = glob.glob("dataset_cache/*.nc")
+        s2_files = [f for f in cache_files if len(os.path.basename(f)) == 35] # S2 cache filenames usually have length 32 + 3 (.nc)
+        
+        if not s2_files:
+            print("[WARNING] Không tìm thấy dữ liệu S2 trong cache! Dùng dummy data.")
+            self._create_dummy()
+            return
+            
+        try:
+            print(f"[DATA] Loading real data from {s2_files[0]}")
+            ds = xr.open_dataset(s2_files[0], engine='netcdf4')
+            if 'time' not in ds.dims or len(ds.time) < sequence_length + 1:
+                self._create_dummy()
+                return
+                
+            # Tính NDVI: (B08 - B04) / (B08 + B04)
+            b8 = ds['B08'].astype(np.float32)
+            b4 = ds['B04'].astype(np.float32)
+            ndvi = (b8 - b4) / (b8 + b4 + 1e-8)
+            ndvi = ndvi.fillna(0).values # shape: (time, y, x)
+            
+            # Lấy 1 pixel trung tâm hoặc toàn bộ ảnh
+            if not self.spatial:
+                # Average pooling over space for 1D time series
+                ndvi = ndvi.mean(axis=(1, 2)) # shape: (time,)
+                for i in range(len(ndvi) - sequence_length):
+                    self.data_seqs.append(ndvi[i:i+sequence_length])
+                    self.targets.append(ndvi[i+sequence_length])
+            else:
+                # Spatial data for ConvLSTM
+                # Downsample to 64x64 to avoid OOM
+                from skimage.transform import resize
+                T = len(ndvi)
+                ndvi_resized = np.zeros((T, 64, 64))
+                for t in range(T):
+                    ndvi_resized[t] = resize(ndvi[t], (64, 64))
+                    
+                for i in range(T - sequence_length):
+                    self.data_seqs.append(ndvi_resized[i:i+sequence_length]) # (seq, 64, 64)
+                    self.targets.append(ndvi_resized[i+sequence_length]) # (64, 64)
+                    
+        except Exception as e:
+            print(f"[ERROR] {e}. Dùng dummy data.")
+            self._create_dummy()
+
+    def _create_dummy(self):
+        T = 20
+        if not self.spatial:
+            ndvi = np.random.rand(T).astype(np.float32)
+            for i in range(T - self.sequence_length):
+                self.data_seqs.append(ndvi[i:i+self.sequence_length])
+                self.targets.append(ndvi[i+self.sequence_length])
+        else:
+            ndvi = np.random.rand(T, 64, 64).astype(np.float32)
+            for i in range(T - self.sequence_length):
+                self.data_seqs.append(ndvi[i:i+self.sequence_length])
+                self.targets.append(ndvi[i+self.sequence_length])
+                
+    def __len__(self):
+        return len(self.data_seqs)
+
+    def __getitem__(self, idx):
+        x = torch.tensor(self.data_seqs[idx], dtype=torch.float32)
+        y = torch.tensor(self.targets[idx], dtype=torch.float32)
+        
+        if not self.spatial:
+            x = x.unsqueeze(1) # (seq_len, features=1)
+            y = y.unsqueeze(0) # (1,)
+        else:
+            x = x.unsqueeze(1) # (seq_len, channels=1, H, W)
+            y = y.unsqueeze(0) # (1, H, W)
+            
+        return x, y