gaia.training.unified_trainer module

GAIA Unified Trainer

This module provides a unified training system that integrates all GAIA components: - Fuzzy simplicial sets for data encoding - Coalgebras for parameter evolution - Hierarchical message passing - Business unit communication - Horn filling and Kan conditions - Endofunctor dynamics

Replaces the massive trainer.py files with an optimal integrated approach.

class gaia.training.unified_trainer.GAIATrainingConfig(input_dim=784, hidden_dims=<factory>, output_dim=10, vocab_size=1000, d_model=256, num_heads=4, num_layers=4, seq_len=32, d_ff=1024, max_seq_length=512, learning_rate=0.001, batch_size=32, max_epochs=100, fuzzy_k_neighbors=5, coalgebra_steps=3, message_passing_levels=3, horn_filling_tolerance=1e-06, use_hierarchical_updates=True, use_business_units=True, use_kan_verification=True, verify_coalgebra_dynamics=True, log_level='INFO', checkpoint_dir='checkpoints', log_interval=10)

Bases: object

Configuration for GAIA unified trainer.

input_dim: int = 784

hidden_dims: List[int]

output_dim: int = 10

vocab_size: int = 1000

d_model: int = 256

num_heads: int = 4

num_layers: int = 4

seq_len: int = 32

d_ff: int = 1024

max_seq_length: int = 512

learning_rate: float = 0.001

batch_size: int = 32

max_epochs: int = 100

fuzzy_k_neighbors: int = 5

coalgebra_steps: int = 3

message_passing_levels: int = 3

horn_filling_tolerance: float = 1e-06

use_hierarchical_updates: bool = True

use_business_units: bool = True

use_kan_verification: bool = True

verify_coalgebra_dynamics: bool = True

log_level: str = 'INFO'

checkpoint_dir: str = 'checkpoints'

log_interval: int = 10

__init__(input_dim=784, hidden_dims=<factory>, output_dim=10, vocab_size=1000, d_model=256, num_heads=4, num_layers=4, seq_len=32, d_ff=1024, max_seq_length=512, learning_rate=0.001, batch_size=32, max_epochs=100, fuzzy_k_neighbors=5, coalgebra_steps=3, message_passing_levels=3, horn_filling_tolerance=1e-06, use_hierarchical_updates=True, use_business_units=True, use_kan_verification=True, verify_coalgebra_dynamics=True, log_level='INFO', checkpoint_dir='checkpoints', log_interval=10)

class gaia.training.unified_trainer.FuzzyDataEncoder(config)

Bases: GAIAComponent

Component for encoding data as fuzzy simplicial sets.

__init__(config)

initialize()

Initialize the fuzzy encoder.

update(state)

Update with new data batch.

validate()

Validate encoder state.

class gaia.training.unified_trainer.CoalgebraEvolution(config, model)

Bases: GAIAComponent

Component for coalgebraic parameter evolution.

__init__(config, model)

initialize()

Initialize coalgebras for model parameters.

update(state)

Update coalgebras with parameter evolution.

validate()

Validate coalgebra evolution.

class gaia.training.unified_trainer.HierarchicalCommunication(config, functor)

Bases: GAIAComponent

Component for hierarchical message passing and business unit communication.

__init__(config, functor)

initialize()

Initialize hierarchical communication systems.

update(state)

Update hierarchical communication.

validate()

Validate hierarchical communication.

class gaia.training.unified_trainer.KanVerification(config, functor)

Bases: GAIAComponent

Component for Kan complex verification and horn filling.

__init__(config, functor)

initialize()

Initialize Kan complex verifier.

update(state)

Update Kan verification.

validate()

Validate Kan verification component.

class gaia.training.unified_trainer.GAIATrainer(model, config)

Bases: IntegratedTrainer

Unified trainer integrating all GAIA components.

__init__(model, config)

train_step()

Execute one training step across all components.

validate_step()

Execute validation across all components.

train_epoch(dataloader)

Train for one epoch.

train(train_loader, val_loader=None, num_epochs=None)

Full training loop.

load_checkpoint(checkpoint_path)

Load training checkpoint.

add_component(component)

Add a component to the trainer.

get_component(name)

Get a component by name.

gaia.training.unified_trainer.create_gaia_trainer(model, config=None)

Create a GAIA unified trainer with default configuration.