bead 0.1.0__py3-none-any.whl

bead/active_learning/models/ordinal_scale.py

@@ -0,0 +1,811 @@
+"""Ordinal scale model for ordered rating scales (Likert, sliders, etc.).
+
+Implements truncated normal distribution for bounded continuous responses on [0, 1].
+Supports GLMM with participant-level random effects (intercepts and slopes).
+"""
+
+from __future__ import annotations
+
+import json
+import tempfile
+from pathlib import Path
+
+import numpy as np
+import torch
+import torch.nn as nn
+from torch.distributions import Normal
+from transformers import AutoModel, AutoTokenizer, TrainingArguments
+
+from bead.active_learning.config import MixedEffectsConfig, VarianceComponents
+from bead.active_learning.models.base import ActiveLearningModel, ModelPrediction
+from bead.active_learning.models.random_effects import RandomEffectsManager
+from bead.active_learning.trainers.data_collator import MixedEffectsDataCollator
+from bead.active_learning.trainers.dataset_utils import items_to_dataset
+from bead.active_learning.trainers.metrics import compute_regression_metrics
+from bead.active_learning.trainers.model_wrapper import EncoderRegressionWrapper
+from bead.config.active_learning import OrdinalScaleModelConfig
+from bead.items.item import Item
+from bead.items.item_template import ItemTemplate, TaskType
+
+__all__ = ["OrdinalScaleModel"]
+
+
+class OrdinalScaleModel(ActiveLearningModel):
+    """Model for ordinal_scale tasks with bounded continuous responses.
+
+    Uses truncated normal distribution on [scale_min, scale_max] to model
+    slider/Likert responses while properly handling endpoints (0 and 1).
+    Supports three modes: fixed effects, random intercepts, random slopes.
+
+    Parameters
+    ----------
+    config : OrdinalScaleModelConfig
+        Configuration object containing all model parameters.
+
+    Attributes
+    ----------
+    config : OrdinalScaleModelConfig
+        Model configuration.
+    tokenizer : AutoTokenizer
+        Transformer tokenizer.
+    encoder : AutoModel
+        Transformer encoder model.
+    regression_head : nn.Sequential
+        Regression head (fixed effects head) - outputs continuous μ.
+    random_effects : RandomEffectsManager
+        Manager for participant-level random effects.
+    variance_history : list[VarianceComponents]
+        Variance component estimates over training (for diagnostics).
+    _is_fitted : bool
+        Whether model has been trained.
+
+    Examples
+    --------
+    >>> from uuid import uuid4
+    >>> from bead.items.item import Item
+    >>> from bead.config.active_learning import OrdinalScaleModelConfig
+    >>> items = [
+    ...     Item(
+    ...         item_template_id=uuid4(),
+    ...         rendered_elements={"text": f"Sentence {i}"}
+    ...     )
+    ...     for i in range(10)
+    ... ]
+    >>> labels = ["0.3", "0.7"] * 5  # Continuous values as strings
+    >>> config = OrdinalScaleModelConfig(  # doctest: +SKIP
+    ...     num_epochs=1, batch_size=2, device="cpu"
+    ... )
+    >>> model = OrdinalScaleModel(config=config)  # doctest: +SKIP
+    >>> metrics = model.train(items, labels, participant_ids=None)  # doctest: +SKIP
+    >>> predictions = model.predict(items[:3], participant_ids=None)  # doctest: +SKIP
+    """
+
+    def __init__(
+        self,
+        config: OrdinalScaleModelConfig | None = None,
+    ) -> None:
+        """Initialize ordinal scale model.
+
+        Parameters
+        ----------
+        config : OrdinalScaleModelConfig | None
+            Configuration object. If None, uses default configuration.
+        """
+        self.config = config or OrdinalScaleModelConfig()
+
+        # Validate mixed_effects configuration
+        super().__init__(self.config)
+
+        self.tokenizer = AutoTokenizer.from_pretrained(self.config.model_name)
+        self.encoder = AutoModel.from_pretrained(self.config.model_name)
+
+        self.regression_head: nn.Sequential | None = None
+        self._is_fitted = False
+
+        # Initialize random effects manager
+        self.random_effects: RandomEffectsManager | None = None
+        self.variance_history: list[VarianceComponents] = []
+
+        self.encoder.to(self.config.device)
+
+    @property
+    def supported_task_types(self) -> list[TaskType]:
+        """Get supported task types.
+
+        Returns
+        -------
+        list[TaskType]
+            List containing "ordinal_scale".
+        """
+        return ["ordinal_scale"]
+
+    def validate_item_compatibility(
+        self, item: Item, item_template: ItemTemplate
+    ) -> None:
+        """Validate item is compatible with ordinal scale model.
+
+        Parameters
+        ----------
+        item : Item
+            Item to validate.
+        item_template : ItemTemplate
+            Template the item was constructed from.
+
+        Raises
+        ------
+        ValueError
+            If task_type is not "ordinal_scale".
+        """
+        if item_template.task_type != "ordinal_scale":
+            raise ValueError(
+                f"Expected task_type 'ordinal_scale', got '{item_template.task_type}'"
+            )
+
+    def _initialize_regression_head(self) -> None:
+        """Initialize regression head for continuous output μ."""
+        hidden_size = self.encoder.config.hidden_size
+
+        # Single output for location parameter μ
+        self.regression_head = nn.Sequential(
+            nn.Linear(hidden_size, 256),
+            nn.ReLU(),
+            nn.Dropout(0.1),
+            nn.Linear(256, 1),  # Output μ (location parameter)
+        )
+        self.regression_head.to(self.config.device)
+
+    def _encode_texts(self, texts: list[str]) -> torch.Tensor:
+        """Encode texts using transformer.
+
+        Parameters
+        ----------
+        texts : list[str]
+            Texts to encode.
+
+        Returns
+        -------
+        torch.Tensor
+            Encoded representations of shape (batch_size, hidden_size).
+        """
+        encodings = self.tokenizer(
+            texts,
+            padding=True,
+            truncation=True,
+            max_length=self.config.max_length,
+            return_tensors="pt",
+        )
+        encodings = {k: v.to(self.config.device) for k, v in encodings.items()}
+
+        outputs = self.encoder(**encodings)
+        return outputs.last_hidden_state[:, 0, :]
+
+    def _prepare_inputs(self, items: list[Item]) -> torch.Tensor:
+        """Prepare inputs for encoding.
+
+        For ordinal scale tasks, concatenates all rendered elements.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Items to encode.
+
+        Returns
+        -------
+        torch.Tensor
+            Encoded representations.
+        """
+        texts = []
+        for item in items:
+            # Concatenate all rendered elements
+            all_text = " ".join(item.rendered_elements.values())
+            texts.append(all_text)
+        return self._encode_texts(texts)
+
+    def _truncated_normal_log_prob(
+        self, y: torch.Tensor, mu: torch.Tensor, sigma: float
+    ) -> torch.Tensor:
+        """Compute log probability of truncated normal distribution.
+
+        Uses truncated normal on [scale_min, scale_max] to properly handle
+        endpoint responses (0.0 and 1.0) without arbitrary nudging.
+
+        Parameters
+        ----------
+        y : torch.Tensor
+            Observed values, shape (batch,).
+        mu : torch.Tensor
+            Location parameters (before truncation), shape (batch,).
+        sigma : float
+            Scale parameter (standard deviation).
+
+        Returns
+        -------
+        torch.Tensor
+            Log probabilities, shape (batch,).
+        """
+        base_dist = Normal(mu.squeeze(), sigma)
+
+        # Unnormalized log prob
+        log_prob_unnorm = base_dist.log_prob(y)
+
+        # Normalizer: log(Φ((high-μ)/σ) - Φ((low-μ)/σ))
+        alpha = (self.config.scale.min - mu.squeeze()) / sigma
+        beta = (self.config.scale.max - mu.squeeze()) / sigma
+        normalizer = base_dist.cdf(beta) - base_dist.cdf(alpha)
+
+        # Clamp to avoid log(0)
+        normalizer = torch.clamp(normalizer, min=1e-8)
+        log_normalizer = torch.log(normalizer)
+
+        return log_prob_unnorm - log_normalizer
+
+    def _prepare_training_data(
+        self,
+        items: list[Item],
+        labels: list[str],
+        participant_ids: list[str],
+        validation_items: list[Item] | None,
+        validation_labels: list[str] | None,
+    ) -> tuple[
+        list[Item], list[float], list[str], list[Item] | None, list[float] | None
+    ]:
+        """Prepare training data for ordinal scale model.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Training items.
+        labels : list[str]
+            Training labels (continuous values as strings).
+        participant_ids : list[str]
+            Normalized participant IDs.
+        validation_items : list[Item] | None
+            Validation items.
+        validation_labels : list[str] | None
+            Validation labels.
+
+        Returns
+        -------
+        tuple[list[Item], list[float], list[str], list[Item] | None, list[float] | None]
+            Prepared items, numeric labels (floats), participant_ids,
+            validation_items, numeric validation_labels.
+        """
+        # Parse labels to floats and validate bounds
+        try:
+            y_values = [float(label) for label in labels]
+        except ValueError as e:
+            raise ValueError(
+                f"Labels must be numeric strings (e.g., '0.5', '0.75'). Got error: {e}"
+            ) from e
+
+        # Validate all values are within bounds
+        for i, val in enumerate(y_values):
+            if not (self.config.scale.min <= val <= self.config.scale.max):
+                raise ValueError(
+                    f"Label at index {i} ({val}) is outside bounds "
+                    f"[{self.config.scale.min}, {self.config.scale.max}]"
+                )
+
+        self._initialize_regression_head()
+
+        # Convert validation labels if provided
+        val_y_numeric = None
+        if validation_items is not None and validation_labels is not None:
+            try:
+                val_y_numeric = [float(label) for label in validation_labels]
+            except ValueError as e:
+                raise ValueError(
+                    f"Validation labels must be numeric strings. Got error: {e}"
+                ) from e
+
+            # Validate bounds for validation labels
+            for i, val in enumerate(val_y_numeric):
+                if not (self.config.scale.min <= val <= self.config.scale.max):
+                    raise ValueError(
+                        f"Validation label at index {i} ({val}) is outside bounds "
+                        f"[{self.config.scale.min}, {self.config.scale.max}]"
+                    )
+
+        return items, y_values, participant_ids, validation_items, val_y_numeric
+
+    def _initialize_random_effects(self, n_classes: int) -> None:
+        """Initialize random effects manager.
+
+        Parameters
+        ----------
+        n_classes : int
+            Number of classes (1 for regression).
+        """
+        self.random_effects = RandomEffectsManager(
+            self.config.mixed_effects,
+            n_classes=n_classes,  # Scalar bias for μ
+        )
+
+    def _do_training(
+        self,
+        items: list[Item],
+        labels_numeric: list[float],
+        participant_ids: list[str],
+        validation_items: list[Item] | None,
+        validation_labels_numeric: list[float] | None,
+    ) -> dict[str, float]:
+        """Perform ordinal scale model training.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Training items.
+        labels_numeric : list[float]
+            Numeric labels (continuous values).
+        participant_ids : list[str]
+            Participant IDs.
+        validation_items : list[Item] | None
+            Validation items.
+        validation_labels_numeric : list[float] | None
+            Numeric validation labels.
+
+        Returns
+        -------
+        dict[str, float]
+            Training metrics.
+        """
+        # Convert validation_labels_numeric back to string labels for validation metrics
+        validation_labels = None
+        if validation_items is not None and validation_labels_numeric is not None:
+            validation_labels = [str(val) for val in validation_labels_numeric]
+
+        # Use HuggingFace Trainer for fixed and random_intercepts modes
+        # random_slopes requires custom loop due to per-participant heads
+        use_huggingface_trainer = self.config.mixed_effects.mode in (
+            "fixed",
+            "random_intercepts",
+        )
+
+        if use_huggingface_trainer:
+            metrics = self._train_with_huggingface_trainer(
+                items,
+                labels_numeric,
+                participant_ids,
+                validation_items,
+                validation_labels,
+            )
+        else:
+            # Use custom training loop for random_slopes
+            metrics = self._train_with_custom_loop(
+                items,
+                labels_numeric,
+                participant_ids,
+                validation_items,
+                validation_labels,
+            )
+
+        # Add validation MSE if validation data provided and not already computed
+        if (
+            validation_items is not None
+            and validation_labels is not None
+            and "val_mse" not in metrics
+        ):
+            # Validation with placeholder participant_ids for mixed effects
+            if self.config.mixed_effects.mode == "fixed":
+                val_participant_ids = ["_fixed_"] * len(validation_items)
+            else:
+                val_participant_ids = ["_validation_"] * len(validation_items)
+            val_predictions = self._do_predict(validation_items, val_participant_ids)
+            val_pred_values = [float(p.predicted_class) for p in val_predictions]
+            val_true_values = [float(label) for label in validation_labels]
+            val_mse = np.mean(
+                [
+                    (pred - true) ** 2
+                    for pred, true in zip(val_pred_values, val_true_values, strict=True)
+                ]
+            )
+            metrics["val_mse"] = val_mse
+
+        return metrics
+
+    def _train_with_huggingface_trainer(
+        self,
+        items: list[Item],
+        y_numeric: list[float],
+        participant_ids: list[str],
+        validation_items: list[Item] | None,
+        validation_labels: list[str] | None,
+    ) -> dict[str, float]:
+        """Train using HuggingFace Trainer with mixed effects support for regression.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Training items.
+        y_numeric : list[float]
+            Numeric labels (continuous values).
+        participant_ids : list[str]
+            Participant IDs.
+        validation_items : list[Item] | None
+            Validation items.
+        validation_labels : list[str] | None
+            Validation labels.
+
+        Returns
+        -------
+        dict[str, float]
+            Training metrics.
+        """
+        # Convert items to HuggingFace Dataset
+        train_dataset = items_to_dataset(
+            items=items,
+            labels=y_numeric,
+            participant_ids=participant_ids,
+            tokenizer=self.tokenizer,
+            max_length=self.config.max_length,
+        )
+
+        eval_dataset = None
+        if validation_items is not None and validation_labels is not None:
+            val_y_numeric = [float(label) for label in validation_labels]
+            val_participant_ids = (
+                ["_validation_"] * len(validation_items)
+                if self.config.mixed_effects.mode != "fixed"
+                else ["_fixed_"] * len(validation_items)
+            )
+            eval_dataset = items_to_dataset(
+                items=validation_items,
+                labels=val_y_numeric,
+                participant_ids=val_participant_ids,
+                tokenizer=self.tokenizer,
+                max_length=self.config.max_length,
+            )
+
+        # Wrap the encoder and regression head for Trainer
+        wrapped_model = EncoderRegressionWrapper(
+            encoder=self.encoder, regression_head=self.regression_head
+        )
+
+        # Create data collator
+        data_collator = MixedEffectsDataCollator(tokenizer=self.tokenizer)
+
+        # Create training arguments with checkpointing
+        with tempfile.TemporaryDirectory() as tmpdir:
+            checkpoint_dir = Path(tmpdir) / "checkpoints"
+            checkpoint_dir.mkdir(parents=True, exist_ok=True)
+
+            training_args = TrainingArguments(
+                output_dir=str(checkpoint_dir),
+                num_train_epochs=self.config.num_epochs,
+                per_device_train_batch_size=self.config.batch_size,
+                per_device_eval_batch_size=self.config.batch_size,
+                learning_rate=self.config.learning_rate,
+                logging_steps=10,
+                eval_strategy="epoch" if eval_dataset is not None else "no",
+                save_strategy="epoch",
+                save_total_limit=1,
+                load_best_model_at_end=False,
+                report_to="none",
+                remove_unused_columns=False,
+                use_cpu=self.config.device == "cpu",
+            )
+
+            # Import here to avoid circular import
+            from bead.active_learning.trainers.mixed_effects import (  # noqa: PLC0415
+                MixedEffectsTrainer,
+            )
+
+            # Create trainer
+            trainer = MixedEffectsTrainer(
+                model=wrapped_model,
+                args=training_args,
+                train_dataset=train_dataset,
+                eval_dataset=eval_dataset,
+                data_collator=data_collator,
+                tokenizer=self.tokenizer,
+                random_effects_manager=self.random_effects,
+                compute_metrics=compute_regression_metrics,
+            )
+
+            # Train
+            train_result = trainer.train()
+
+            # Get training metrics
+            train_metrics = trainer.evaluate(eval_dataset=train_dataset)
+            metrics: dict[str, float] = {
+                "train_loss": float(train_result.training_loss),
+                "train_mse": train_metrics.get("eval_mse", 0.0),
+                "train_mae": train_metrics.get("eval_mae", 0.0),
+                "train_r2": train_metrics.get("eval_r2", 0.0),
+            }
+
+            # Get validation metrics if eval_dataset was provided
+            if eval_dataset is not None:
+                val_metrics = trainer.evaluate(eval_dataset=eval_dataset)
+                metrics.update(
+                    {
+                        "val_mse": val_metrics.get("eval_mse", 0.0),
+                        "val_mae": val_metrics.get("eval_mae", 0.0),
+                        "val_r2": val_metrics.get("eval_r2", 0.0),
+                    }
+                )
+
+        return metrics
+
+    def _train_with_custom_loop(
+        self,
+        items: list[Item],
+        y_numeric: list[float],
+        participant_ids: list[str],
+        validation_items: list[Item] | None,
+        validation_labels: list[str] | None,
+    ) -> dict[str, float]:
+        """Train using custom loop for random_slopes mode.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Training items.
+        y_numeric : list[float]
+            Numeric labels (continuous values).
+        participant_ids : list[str]
+            Participant IDs.
+        validation_items : list[Item] | None
+            Validation items.
+        validation_labels : list[str] | None
+            Validation labels.
+
+        Returns
+        -------
+        dict[str, float]
+            Training metrics.
+        """
+        y = torch.tensor(y_numeric, dtype=torch.float, device=self.config.device)
+
+        # Build optimizer parameters
+        params_to_optimize = list(self.encoder.parameters()) + list(
+            self.regression_head.parameters()
+        )
+
+        # Add random effects parameters for random_slopes
+        for head in self.random_effects.slopes.values():
+            params_to_optimize.extend(head.parameters())
+
+        optimizer = torch.optim.AdamW(params_to_optimize, lr=self.config.learning_rate)
+
+        self.encoder.train()
+        self.regression_head.train()
+
+        for _epoch in range(self.config.num_epochs):
+            n_batches = (
+                len(items) + self.config.batch_size - 1
+            ) // self.config.batch_size
+            epoch_loss = 0.0
+            epoch_mse = 0.0
+
+            for i in range(n_batches):
+                start_idx = i * self.config.batch_size
+                end_idx = min(start_idx + self.config.batch_size, len(items))
+
+                batch_items = items[start_idx:end_idx]
+                batch_labels = y[start_idx:end_idx]
+                batch_participant_ids = participant_ids[start_idx:end_idx]
+
+                embeddings = self._prepare_inputs(batch_items)
+
+                # Per-participant head for random_slopes
+                mu_list = []
+                for j, pid in enumerate(batch_participant_ids):
+                    participant_head = self.random_effects.get_slopes(
+                        pid,
+                        fixed_head=self.regression_head,
+                        create_if_missing=True,
+                    )
+                    mu_j = participant_head(embeddings[j : j + 1]).squeeze()
+                    mu_list.append(mu_j)
+                mu = torch.stack(mu_list)
+
+                # Negative log-likelihood of truncated normal
+                log_probs = self._truncated_normal_log_prob(
+                    batch_labels, mu, self.config.sigma
+                )
+                loss_nll = -log_probs.mean()
+
+                # Add prior regularization
+                loss_prior = self.random_effects.compute_prior_loss()
+                loss = loss_nll + loss_prior
+
+                optimizer.zero_grad()
+                loss.backward()
+                optimizer.step()
+
+                epoch_loss += loss.item()
+                # Also track MSE for interpretability
+                mse = ((mu - batch_labels) ** 2).mean().item()
+                epoch_mse += mse
+
+            epoch_loss = epoch_loss / n_batches
+            epoch_mse = epoch_mse / n_batches
+
+        metrics: dict[str, float] = {
+            "train_loss": epoch_loss,
+            "train_mse": epoch_mse,
+        }
+
+        return metrics
+
+    def _do_predict(
+        self, items: list[Item], participant_ids: list[str]
+    ) -> list[ModelPrediction]:
+        """Perform ordinal scale model prediction.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Items to predict.
+        participant_ids : list[str]
+            Normalized participant IDs.
+
+        Returns
+        -------
+        list[ModelPrediction]
+            Predictions with predicted_class as string representation of value.
+        """
+        self.encoder.eval()
+        self.regression_head.eval()
+
+        with torch.no_grad():
+            embeddings = self._prepare_inputs(items)
+
+            # Forward pass depends on mixed effects mode
+            if self.config.mixed_effects.mode == "fixed":
+                mu = self.regression_head(embeddings).squeeze(1)
+
+            elif self.config.mixed_effects.mode == "random_intercepts":
+                mu = self.regression_head(embeddings).squeeze(1)
+                for i, pid in enumerate(participant_ids):
+                    # Unknown participants: use prior mean (zero bias)
+                    bias = self.random_effects.get_intercepts(
+                        pid, n_classes=1, param_name="mu", create_if_missing=False
+                    )
+                    mu[i] = mu[i] + bias.item()
+
+            elif self.config.mixed_effects.mode == "random_slopes":
+                mu_list = []
+                for i, pid in enumerate(participant_ids):
+                    # Unknown participants: use fixed head
+                    participant_head = self.random_effects.get_slopes(
+                        pid, fixed_head=self.regression_head, create_if_missing=False
+                    )
+                    mu_i = participant_head(embeddings[i : i + 1]).squeeze()
+                    mu_list.append(mu_i)
+                mu = torch.stack(mu_list)
+
+            # Clamp predictions to bounds
+            mu = torch.clamp(mu, self.config.scale.min, self.config.scale.max)
+            predictions_array = mu.cpu().numpy()
+
+        predictions = []
+        for i, item in enumerate(items):
+            pred_value = float(predictions_array[i])
+            predictions.append(
+                ModelPrediction(
+                    item_id=str(item.id),
+                    probabilities={},  # Not applicable for regression
+                    predicted_class=str(pred_value),  # Continuous value as string
+                    confidence=1.0,  # Not applicable for regression
+                )
+            )
+
+        return predictions
+
+    def _do_predict_proba(
+        self, items: list[Item], participant_ids: list[str]
+    ) -> np.ndarray:
+        """Perform ordinal scale model probability prediction.
+
+        For ordinal scale regression, returns μ values directly.
+
+        Parameters
+        ----------
+        items : list[Item]
+            Items to predict.
+        participant_ids : list[str]
+            Normalized participant IDs.
+
+        Returns
+        -------
+        np.ndarray
+            Array of shape (n_items, 1) with predicted μ values.
+        """
+        predictions = self._do_predict(items, participant_ids)
+        return np.array([[float(p.predicted_class)] for p in predictions])
+
+    def _save_model_components(self, save_path: Path) -> None:
+        """Save model-specific components.
+
+        Parameters
+        ----------
+        save_path : Path
+            Directory to save to.
+        """
+        self.encoder.save_pretrained(save_path / "encoder")
+        self.tokenizer.save_pretrained(save_path / "encoder")
+
+        torch.save(
+            self.regression_head.state_dict(),
+            save_path / "regression_head.pt",
+        )
+
+    def _get_save_state(self) -> dict[str, object]:
+        """Get model-specific state to save.
+
+        Returns
+        -------
+        dict[str, object]
+            State dictionary.
+        """
+        return {}
+
+    def _restore_training_state(self, config_dict: dict[str, object]) -> None:
+        """Restore model-specific training state.
+
+        Parameters
+        ----------
+        config_dict : dict[str, object]
+            Configuration dictionary with training state.
+        """
+        # OrdinalScaleModel doesn't have additional training state to restore
+        pass
+
+    def _load_model_components(self, load_path: Path) -> None:
+        """Load model-specific components.
+
+        Parameters
+        ----------
+        load_path : Path
+            Directory to load from.
+        """
+        # Load config.json to reconstruct config
+        with open(load_path / "config.json") as f:
+            config_dict = json.load(f)
+
+        # Reconstruct MixedEffectsConfig if needed
+        if "mixed_effects" in config_dict and isinstance(
+            config_dict["mixed_effects"], dict
+        ):
+            config_dict["mixed_effects"] = MixedEffectsConfig(
+                **config_dict["mixed_effects"]
+            )
+
+        self.config = OrdinalScaleModelConfig(**config_dict)
+
+        self.encoder = AutoModel.from_pretrained(load_path / "encoder")
+        self.tokenizer = AutoTokenizer.from_pretrained(load_path / "encoder")
+
+        self._initialize_regression_head()
+        self.regression_head.load_state_dict(
+            torch.load(
+                load_path / "regression_head.pt", map_location=self.config.device
+            )
+        )
+
+        self.encoder.to(self.config.device)
+        self.regression_head.to(self.config.device)
+
+    def _get_n_classes_for_random_effects(self) -> int:
+        """Get the number of classes for initializing RandomEffectsManager.
+
+        For ordinal scale models, this is 1 (scalar bias).
+
+        Returns
+        -------
+        int
+            Always 1 for regression.
+        """
+        return 1
+
+    def _get_random_effects_fixed_head(self) -> torch.nn.Module | None:
+        """Get the fixed head for random effects.
+
+        Returns
+        -------
+        torch.nn.Module | None
+            The regression head, or None if not applicable.
+        """
+        return self.regression_head