vertiport_autonomy.training package
Submodules
vertiport_autonomy.training.curriculum module
Curriculum learning trainer for progressive difficulty training.
- class vertiport_autonomy.training.curriculum.CurriculumTrainer(log_dir: str = 'logs', model_dir: str = 'models')[source]
Bases:
objectCurriculum learning trainer for vertiport autonomy.
- __init__(log_dir: str = 'logs', model_dir: str = 'models')[source]
Initialize the curriculum trainer.
- Parameters:
log_dir – Directory for training logs
model_dir – Directory for saving models
- set_custom_phases(phases: List[Dict[str, Any]]) None[source]
Set custom curriculum phases.
- Parameters:
phases – List of phase configurations
- train_phase(phase_config: Dict[str, Any], model: PPO | None = None) PPO[source]
Train a single curriculum phase.
- Parameters:
phase_config – Configuration for this phase
model – Previous model to continue from (None for first phase)
- Returns:
Trained model for this phase
- run_full_curriculum() PPO[source]
Run the complete curriculum learning process.
- Returns:
Final trained model
- run_single_phase(phase_name: str, model: PPO | None = None) PPO[source]
Run a single phase of the curriculum.
- Parameters:
phase_name – Name of the phase to run
model – Optional model to continue from
- Returns:
Trained model for this phase
- Raises:
ValueError – If phase_name is not found
vertiport_autonomy.training.trainer module
Basic training utilities for vertiport autonomy agents.
- class vertiport_autonomy.training.trainer.Trainer(log_dir: str = 'logs', model_dir: str = 'models', n_envs: int = 50, **ppo_kwargs)[source]
Bases:
objectBasic trainer for PPO agents in vertiport environments.
- __init__(log_dir: str = 'logs', model_dir: str = 'models', n_envs: int = 50, **ppo_kwargs)[source]
Initialize the trainer.
- Parameters:
log_dir – Directory for training logs
model_dir – Directory for saving models
n_envs – Number of parallel environments
**ppo_kwargs – Additional arguments for PPO
- create_environment(scenario_path: str) VecNormalize[source]
Create a vectorized and normalized environment.
- Parameters:
scenario_path – Path to scenario configuration file
- Returns:
Normalized vectorized environment
- create_model(env: VecNormalize, **override_params) PPO[source]
Create a PPO model.
- Parameters:
env – Environment for training
**override_params – Parameters to override defaults
- Returns:
PPO model instance
- create_callbacks(save_freq: int = 50000, eval_freq: int = 10000, n_eval_episodes: int = 5, name_prefix: str = 'ppo_vertiport') list[source]
Create training callbacks.
- Parameters:
save_freq – Frequency for saving checkpoints
eval_freq – Frequency for evaluation
n_eval_episodes – Number of episodes for evaluation
name_prefix – Prefix for saved model names
- Returns:
List of callbacks
- train(scenario_path: str, total_timesteps: int, tb_log_name: str = 'PPO_Vertiport', save_final: bool = True, final_model_name: str = 'ppo_vertiport_final', **model_params) PPO[source]
Train a PPO agent.
- Parameters:
scenario_path – Path to scenario configuration
total_timesteps – Total training timesteps
tb_log_name – TensorBoard log name
save_final – Whether to save final model
final_model_name – Name for final model
**model_params – Additional model parameters
- Returns:
Trained PPO model
Module contents
Training utilities and frameworks.
- class vertiport_autonomy.training.Trainer(log_dir: str = 'logs', model_dir: str = 'models', n_envs: int = 50, **ppo_kwargs)[source]
Bases:
objectBasic trainer for PPO agents in vertiport environments.
- __init__(log_dir: str = 'logs', model_dir: str = 'models', n_envs: int = 50, **ppo_kwargs)[source]
Initialize the trainer.
- Parameters:
log_dir – Directory for training logs
model_dir – Directory for saving models
n_envs – Number of parallel environments
**ppo_kwargs – Additional arguments for PPO
- create_callbacks(save_freq: int = 50000, eval_freq: int = 10000, n_eval_episodes: int = 5, name_prefix: str = 'ppo_vertiport') list[source]
Create training callbacks.
- Parameters:
save_freq – Frequency for saving checkpoints
eval_freq – Frequency for evaluation
n_eval_episodes – Number of episodes for evaluation
name_prefix – Prefix for saved model names
- Returns:
List of callbacks
- create_environment(scenario_path: str) VecNormalize[source]
Create a vectorized and normalized environment.
- Parameters:
scenario_path – Path to scenario configuration file
- Returns:
Normalized vectorized environment
- create_model(env: VecNormalize, **override_params) PPO[source]
Create a PPO model.
- Parameters:
env – Environment for training
**override_params – Parameters to override defaults
- Returns:
PPO model instance
- train(scenario_path: str, total_timesteps: int, tb_log_name: str = 'PPO_Vertiport', save_final: bool = True, final_model_name: str = 'ppo_vertiport_final', **model_params) PPO[source]
Train a PPO agent.
- Parameters:
scenario_path – Path to scenario configuration
total_timesteps – Total training timesteps
tb_log_name – TensorBoard log name
save_final – Whether to save final model
final_model_name – Name for final model
**model_params – Additional model parameters
- Returns:
Trained PPO model
- class vertiport_autonomy.training.CurriculumTrainer(log_dir: str = 'logs', model_dir: str = 'models')[source]
Bases:
objectCurriculum learning trainer for vertiport autonomy.
- __init__(log_dir: str = 'logs', model_dir: str = 'models')[source]
Initialize the curriculum trainer.
- Parameters:
log_dir – Directory for training logs
model_dir – Directory for saving models
- get_phase_names() List[str][source]
Get list of available phase names.
- Returns:
List of phase names
- run_full_curriculum() PPO[source]
Run the complete curriculum learning process.
- Returns:
Final trained model
- run_single_phase(phase_name: str, model: PPO | None = None) PPO[source]
Run a single phase of the curriculum.
- Parameters:
phase_name – Name of the phase to run
model – Optional model to continue from
- Returns:
Trained model for this phase
- Raises:
ValueError – If phase_name is not found