Launch

The launch function is the entry point for starting a PAMIQ-Core system. It initializes all components, connects them together, and manages the system's lifecycle.

Breaking Changes in v0.5

The launch function parameter data has been renamed to buffers for better clarity. If you're upgrading from v0.4, update your code:

# Before (v0.4)
launch(
    interaction=interaction,
    models=models,
    data={"buffer": my_buffer},  # Old parameter name
    trainers=trainers,
    config=config
)

# After (v0.5)
launch(
    interaction=interaction,
    models=models,
    buffers={"buffer": my_buffer},  # New parameter name
    trainers=trainers,
    config=config
)

Basic Usage

To launch a PAMIQ-Core system, you need to provide your interaction components, models, data buffers, trainers, and configuration settings:

from pamiq_core import launch, LaunchConfig, Interaction

# Create your agent and environment
agent = YourAgent()
environment = YourEnvironment()

# Create an interaction between them
interaction = Interaction(agent, environment)

# Launch the system
launch(
    interaction=interaction,
    models={"model_name": your_model},
    buffers={"buffer_name": your_data_buffer},
    trainers={"trainer_name": your_trainer},
    config=LaunchConfig(
        states_dir="./saved_states",
        max_uptime=3600,  # Run for 1 hour
        time_scale=2.0    # Run at 2x speed
    )
)

Common Configuration Scenarios

Accelerated Learning

To speed up time for faster training:

config = LaunchConfig(
    time_scale=10.0  # Run at 10x speed
)

Disabling Web API

To run the system without remote control capabilities:

config = LaunchConfig(
    web_api_address=None,  # Disable web API server
)

Resumable Training

To save system state for later resumption:

from pamiq_core.state_persistence import PeriodicSaveCondition, LatestStatesKeeper

# Initial run with automatic state management
launch(
    interaction=interaction,
    models=models,
    buffers=buffers,
    trainers=trainers,
    config=LaunchConfig(
        states_dir="./saved_states",
        save_state_condition=PeriodicSaveCondition(600.0),  # Save every 10 minutes
        states_keeper=LatestStatesKeeper(
            states_dir="./saved_states",
            max_keep=5  # Keep only the 5 most recent states
        )
    )
)

# Later, resume from the last saved state
latest_state = list(Path("./saved_states").glob("*.state"))[-1]
launch(
    interaction=interaction,
    models=models,
    buffers=buffers,
    trainers=trainers,
    config=LaunchConfig(
        states_dir="./saved_states",
        saved_state_path=latest_state
    )
)

API Reference

More details, Checkout to the API Reference