Upload configuration_mirai.py with huggingface_hub

configuration_mirai.py

@@ -0,0 +1,130 @@
+"""Configuration class for MIRAI model."""
+
+from transformers import PretrainedConfig
+from typing import List, Dict, Any, Optional
+
+
+class MiraiConfig(PretrainedConfig):
+    """
+    Configuration class for MIRAI breast cancer risk prediction model.
+
+    Args:
+        num_classes: Number of prediction classes (default: 5 for 5-year predictions)
+        img_size: Input image size [height, width] (default: [1664, 2048])
+        num_chan: Number of image channels (default: 3)
+        num_images: Number of mammogram views (default: 4)
+        multi_image: Whether to use multiple images (default: True)
+        encoder_config: Configuration for the ResNet encoder
+        transformer_config: Configuration for the transformer module
+        risk_factors: Configuration for clinical risk factors
+        preprocessing: Image preprocessing parameters
+        **kwargs: Additional configuration parameters
+    """
+
+    model_type = "mirai"
+
+    def __init__(
+        self,
+        num_classes: int = 5,
+        img_size: List[int] = None,
+        num_chan: int = 3,
+        num_images: int = 4,
+        multi_image: bool = True,
+        encoder_config: Dict[str, Any] = None,
+        transformer_config: Dict[str, Any] = None,
+        risk_factors: Dict[str, Any] = None,
+        preprocessing: Dict[str, Any] = None,
+        model_version: str = "1.0.0",
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+
+        # Model architecture
+        self.num_classes = num_classes
+        self.img_size = img_size or [1664, 2048]
+        self.num_chan = num_chan
+        self.num_images = num_images
+        self.multi_image = multi_image
+        self.model_version = model_version
+
+        # Encoder configuration
+        self.encoder_config = encoder_config or {
+            "architecture": "resnet",
+            "block_layout": [
+                ["BasicBlock", 2],
+                ["BasicBlock", 2],
+                ["BasicBlock", 2],
+                ["BasicBlock", 2]
+            ],
+            "pool_name": "GlobalMaxPool",
+            "img_only_dim": 2048,
+            "dropout": 0.25,
+            "pretrained_on_imagenet": False
+        }
+
+        # Transformer configuration
+        self.transformer_config = transformer_config or {
+            "hidden_dim": 256,
+            "num_heads": 8,
+            "num_layers": 6,
+            "dropout": 0.1,
+            "max_seq_length": 4
+        }
+
+        # Risk factors configuration
+        self.risk_factors = risk_factors or {
+            "use_risk_factors": True,
+            "num_risk_factors": 34,
+            "risk_factor_keys": [
+                "density", "binary_family_history", "binary_biopsy_benign", "binary_biopsy_LCIS",
+                "binary_biopsy_atypical_hyperplasia", "age", "menarche_age", "menopause_age",
+                "first_pregnancy_age", "prior_hist", "race", "parous", "menopausal_status",
+                "weight", "height", "ovarian_cancer", "ovarian_cancer_age", "ashkenazi",
+                "brca", "mom_bc_cancer_history", "m_aunt_bc_cancer_history",
+                "p_aunt_bc_cancer_history", "m_grandmother_bc_cancer_history",
+                "p_grantmother_bc_cancer_history", "sister_bc_cancer_history",
+                "mom_oc_cancer_history", "m_aunt_oc_cancer_history",
+                "p_aunt_oc_cancer_history", "m_grandmother_oc_cancer_history",
+                "p_grantmother_oc_cancer_history", "sister_oc_cancer_history",
+                "hrt_type", "hrt_duration", "hrt_years_ago_stopped"
+            ]
+        }
+
+        # Preprocessing configuration
+        self.preprocessing = preprocessing or {
+            "img_mean": 7047.99,
+            "img_std": 12005.5,
+            "normalize_method": "imagenet",
+            "imagenet_mean": [0.485, 0.456, 0.406],
+            "imagenet_std": [0.229, 0.224, 0.225]
+        }
+
+    @property
+    def use_risk_factors(self) -> bool:
+        """Whether the model uses clinical risk factors."""
+        return self.risk_factors.get("use_risk_factors", True)
+
+    @property
+    def num_risk_factors(self) -> int:
+        """Number of clinical risk factors."""
+        return self.risk_factors.get("num_risk_factors", 34)
+
+    @property
+    def risk_factor_keys(self) -> List[str]:
+        """List of risk factor keys."""
+        return self.risk_factors.get("risk_factor_keys", [])
+
+    @property
+    def img_height(self) -> int:
+        """Image height."""
+        return self.img_size[0]
+
+    @property
+    def img_width(self) -> int:
+        """Image width."""
+        return self.img_size[1]
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert configuration to dictionary."""
+        output = super().to_dict()
+        return output