Create config.py

config.py

@@ -0,0 +1,81 @@
+"""
+Configuration module for biomass prediction model.
+This recreates the config class needed for unpickling the model package.
+
+Author: najahpokkiri
+Date: 2025-05-17
+"""
+from dataclasses import dataclass, field
+from typing import List, Dict, Optional, Union, Tuple
+from datetime import datetime
+
+@dataclass
+class BiomassPipelineConfig:
+    """Configuration for Biomass Prediction Pipeline"""
+    # Core settings
+    mode: str = "full"
+    random_seed: int = 42
+    project_name: str = "biomass-prediction"
+    created_by: str = "najahpokkiri"
+    created_date: str = "2025-05-16"
+    
+    # Data paths
+    raster_pairs: List[tuple] = field(default_factory=lambda: [
+        ("/teamspace/studios/dl2/clean/data/s1_s2_l8_palsar_ch_dem_yellapur_2020.tif",
+         "/teamspace/studios/dl2/clean/data/agbd_yellapur_reprojected_1.tif"),
+        ("/teamspace/studios/dl2/clean/data/s1_s2_l8_palsar_ch_dem_betul_2020.tif",
+         "/teamspace/studios/dl2/clean/data/agbd_betul_reprojected_1.tif"),
+        ("/teamspace/studios/dl2/clean/data/s1_s2_l8_palsar_ch_goa_achankumar_2020_clipped.tif",
+         "/teamspace/studios/dl2/clean/data/02_Achanakmar_AGB40_band1_onImgGrid.tif"),
+        ("/teamspace/studios/dl2/clean/data/s1_s2_l8_palsar_ch_dem_Khaoyai_2020_clipped.tif",
+         "/teamspace/studios/dl2/clean/data/05_Khaoyai_AGB40_band1_onImgGrid.tif")
+    ])
+    data_dir: str = "data/"
+    results_dir: str = field(default_factory=lambda: f"biomass_results_{datetime.now().strftime('%Y%m%d_%H%M%S')}")
+    
+    # Feature engineering settings
+    use_log_transform: bool = True
+    epsilon: float = 1.0
+    use_advanced_indices: bool = True
+    use_texture_features: bool = True
+    use_spatial_features: bool = True
+    use_pca_features: bool = True
+    pca_components: int = 25
+    scale_method: str = "robust"
+    outlier_removal: bool = True
+    outlier_threshold: float = 3.0
+    
+    # Model settings
+    model_type: str = "StableResNet"
+    dropout_rate: float = 0.2
+    batch_size: int = 256
+    learning_rate: float = 0.001
+    weight_decay: float = 1e-5
+    max_epochs: int = 300
+    patience: int = 30
+    test_size: float = 0.15
+    val_size: float = 0.15
+    
+    # Deployment settings
+    huggingface_repo: str = "najahpokkiri/biomass-prediction"
+    quantize_model: bool = False
+    max_samples_per_tile: Optional[int] = None
+    
+    def __post_init__(self):
+        """Adjust settings based on mode"""
+        if self.mode == "test":
+            # Quick testing settings
+            self.raster_pairs = self.raster_pairs[:1]  # Use only first tile
+            self.max_epochs = 10
+            self.batch_size = 32
+            self.use_texture_features = False
+            self.use_spatial_features = False
+            self.use_pca_features = False
+            self.max_samples_per_tile = 5000
+            self.pca_components = 10
+        else:
+            # Full mode settings
+            self.max_samples_per_tile = None
+
+# Create a default config instance
+default_config = BiomassPipelineConfig()