Configuration¶

SLEAP-NN uses YAML configuration files with three main sections.

Section	Description
Data Config	Data paths, preprocessing, augmentation
Model Config	Backbone and head architecture
Trainer Config	Epochs, optimizer, learning rate, logging
Sample Configs	Ready-to-use templates

Generate Configs¶

Don't want to write YAML by hand? Use these tools to generate configs:

Tool	Description
Config Picker	Interactive web app (runs in browser)
`sleap-nn config`	CLI with interactive TUI or auto mode

See the Config Generator Guide for detailed CLI usage.

Config Structure¶

data_config:      # Data loading and preprocessing
  train_labels_path: [...]
  preprocessing: {...}
  augmentation_config: {...}

model_config:     # Model architecture
  backbone_config: {...}
  head_configs: {...}

trainer_config:   # Training settings
  max_epochs: 100
  optimizer: {...}

Usage¶

sleap-nn train --config config.yaml

Converting from Legacy SLEAP¶

If you have a training_config.json from SLEAP < v1.5, you can convert it to the new YAML format:

from sleap_nn.config.training_job_config import TrainingJobConfig
from omegaconf import OmegaConf

# Load legacy config and convert
config = TrainingJobConfig.load_sleap_config("path/to/training_config.json")

# Save as YAML
OmegaConf.save(config, "config.yaml")