Task

dnallm.tasks.task

DNA Language Model Fine-tuning Task Definition Module.

This module defines various task types and related components supported by DNA language models during fine-tuning, including:

1. TaskType: Task type enumeration
2. Binary classification (BINARY): e.g., promoter prediction, enhancer identification
3. Multi-class classification (MULTICLASS): e.g., protein family classification, functional region classification
4. Regression (REGRESSION): e.g., expression level prediction, binding strength prediction
5. Token classification (NER): Named Entity Recognition tasks

6. Generation and embedding tasks for different model architectures

7. TaskConfig: Task configuration class

8. Configures task type, number of labels, label names, etc.

9. Provides threshold settings for binary classification tasks

10. TaskHead: Task-specific prediction heads

11. Provides specialized neural network layers for different task types
12. Supports feature dimensionality reduction and dropout to prevent overfitting

13. Automatically selects output dimensions based on task type

14. compute_metrics: Evaluation metric computation

15. Binary: accuracy, F1 score
16. Multi-class: accuracy, macro F1, weighted F1
17. Regression: mean squared error, R-squared value

Usage example

task_config = TaskConfig( task_type=TaskType.BINARY, num_labels=2, label_names=["negative", "positive"] )

Classes

TaskConfig

Bases: BaseModel

Configuration class for different fine-tuning tasks.

This class provides a structured way to configure task-specific parameters including task type, number of labels, label names, and classification thresholds.

Attributes:

Name	Type	Description
task_type	str	Type of task to perform (must match regex pattern)
num_labels	int	Number of output labels/classes
label_names	list | None	List of label names for classification tasks
threshold	float	Classification threshold for binary and multi-label tasks

Functions

model_post_init

model_post_init(__context)

Initialize task configuration after model validation.

This method is called after Pydantic model validation and automatically sets appropriate default values based on task type.

TaskType

Bases: Enum

Enum for supported task types in DNALLM.

This enum defines the various tasks that can be performed with DNA language models, including classification, regression, and generation tasks.