Crate rbp_autotrain 

Automated training pipeline orchestration.

This module manages the complete training workflow, from checking database state through clustering and blueprint generation. Supports both single-machine and distributed training modes.

Pipeline Stages

1. Pretraining — Generate abstractions via hierarchical clustering
2. Fast mode — Single-machine MCCFR with in-memory profile
3. Slow mode — Distributed workers with PostgreSQL synchronization

Core Types

• Trainer — Main entry point for training orchestration
• Mode — Training configuration (fast vs slow, clustering vs blueprint)

Submodules

• workers — Distributed training workers for MCCFR

Re-exports

pub use workers::*;

Modules

workers

Distributed training workers for MCCFR.

Structs

EpochMeta

Newtype wrapper for epoch counter (enables Schema implementation).

FastSession

Fast in-memory training using Pluribus.

PreTraining

Zero-sized orchestrator for the clustering pipeline. Encapsulates all clustering logic so Trainer stays clean.

SlowSession

Slow distributed training using Worker pool.

Enums

Mode

Training mode parsed from command line arguments

Traits

Trainer

Unified training session interface. Both fast and slow modes implement this for polymorphic training loops.