id: 39b1f30252c641b0852fd996e7b4ddd8
parent_id: 243ce0c2b827464295e74f126a7d9c26
item_type: 1
item_id: c7a39e16770542a08004fd73e971ca1f
item_updated_time: 1784263384821
title_diff: "[]"
body_diff: "[{\"diffs\":[[0,\"ng\\\n\\\n\"],[-1,\"> **Current state, 2026-07-03. Range-weighted equity features added (280→290 dims). RL self-play loop running with frozen backbone + REINFORCE. Model at iteration ~110 (last successful update before FEATURE_DIM break). Need fresh data collection + retrain with new 290-dim features.**\\\n\\\n## Role in the Framework\\\n\\\nGen 5 replaces the heuristic decision logic (Gen 3/4 thresholds) with a **neural network** trained via imitation learning then RL fine-tuning (REINFORCE). It uses the Gen 4 observer's stats as input features and learns when to trust equity vs exploit opponent tendencies, optimal bet sizing, and bluffing frequencies.\\\n\\\n## Architecture\\\n\\\n### Multi-Head Network (current production)\\\n\\\n```\\\nInput(290) → Linear(512) → ReLU → LayerNorm(512)     [backbone, frozen during RL]\\\n           → Linear(512) → ReLU → LayerNorm(512)     [backbone, frozen during RL]\\\n           → Linear(256) → ReLU                       [backbone, frozen during RL]\\\n           → Action head: Linear(3)  → masked softmax  [passive/call/aggressive]\\\n           → Sizing head: Linear(5)  → softmax          [third/half/2third/pot/allin]\\\n           → Value head:  Linear(1)                     [not used in RL currently]\\\n```\\\n\\\n~1M parameters. Inference <1ms on CPU. LayerNorm bug in Candle 0.8.4 fixed (manual layer_norm replacing broken `ops::layer_norm`).\\\n\\\n### Feature Vector (290 dims)\\\n\\\n| Section | Dims | Content |\\\n|---\"],[0,\"|---\"]],\"start1\":30,\"start2\":30,\"length1\":1403,\"length2\":8}]"
metadata_diff: {"new":{},"deleted":[]}
encryption_cipher_text: 
encryption_applied: 0
updated_time: 2026-07-17T04:46:45.672Z
created_time: 2026-07-17T04:46:45.672Z
type_: 13