feat: implement comprehensive land cover classification pipeline with model benchmarking and experiment logging

analyze_data.py

@@ -0,0 +1,14 @@
+import joblib, numpy as np
+data = joblib.load('dataset_cache/training_data_2d_temporal.joblib')
+X, y = data['X'], data['y']
+print(f'Shape: {X.shape}, dtype: {X.dtype}')
+print(f'Labels unique: {np.unique(y)}')
+print(f'Label counts:')
+for lbl in sorted(np.unique(y)):
+    print(f'  Label {lbl}: {(y==lbl).sum()}')
+print(f'Range: [{X.min():.4f}, {X.max():.4f}], Mean: {X.mean():.4f}')
+print(f'AllZero patches: {(X.reshape(X.shape[0],-1).sum(1)==0).sum()}')
+for t in range(4):
+    block = X[:, t*6:(t+1)*6]
+    nz = (block.reshape(block.shape[0],-1).sum(1)!=0).sum()
+    print(f'  Timestep {t}: non-zero={nz}/{len(X)}')