Files
foxhunt/archive/scripts/test_dqn_initialization.sh
jgrusewski 2df1ea92e1 feat(ml): WAVE 29 DQN Codebase Cleanup & Refactoring Campaign
BREAKING CHANGES:
- Removed orphaned dqn.rs monolithic trainer (4,975 lines)
- Removed orphaned dqn_ensemble.rs module (816 lines)
- Removed orphaned tft.rs and tft_complete_int8_integration_test.rs
- TFT trainer split into modular directory structure

DQN Module Refactoring:
- Split trainers/dqn.rs into modular structure (config.rs, statistics.rs, trainer.rs)
- Fixed hyperopt 39D search space (continuous params only)
- Boolean flags (use_dueling, use_double_dqn, use_per, use_noisy_nets) are now FIXED architectural decisions
- use_distributional defaults to false (Candle BUG #36 - scatter_add gradient issues)

Clean Module Structure:
- ml/src/trainers/dqn/ directory with proper mod.rs exports
- ml/src/trainers/tft/ directory with config.rs, types.rs, model.rs, trainer.rs, tests.rs
- All P0 features validated: TD-error clamping, batch diversity, LR scheduler, priority staleness

Documentation:
- Added comprehensive docs in docs/codebase-cleanup/
- ADR-001 for DQN refactoring decisions
- Rainbow DQN component matrix and quick reference guides

Build Status: Compiles with zero errors

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-27 23:46:13 +01:00

66 lines
2.2 KiB
Bash
Executable File

#!/bin/bash
# Test script to verify DQN initialization is non-deterministic
# Runs 3 parallel training instances and extracts initial Q-values
set -e
echo "=== DQN Non-Deterministic Initialization Test ==="
echo "Starting 3 parallel training runs with 1 epoch each..."
echo ""
# Clean up old test outputs
rm -rf /tmp/init_test_* 2>/dev/null || true
# Run 3 training instances in parallel
cargo run --package ml --example train_dqn --release --features cuda -- \
--epochs 1 --output-dir /tmp/init_test_1 > /tmp/init_test_1.log 2>&1 &
PID1=$!
cargo run --package ml --example train_dqn --release --features cuda -- \
--epochs 1 --output-dir /tmp/init_test_2 > /tmp/init_test_2.log 2>&1 &
PID2=$!
cargo run --package ml --example train_dqn --release --features cuda -- \
--epochs 1 --output-dir /tmp/init_test_3 > /tmp/init_test_3.log 2>&1 &
PID3=$!
echo "Waiting for training runs to complete..."
echo " PID $PID1 (test 1)"
echo " PID $PID2 (test 2)"
echo " PID $PID3 (test 3)"
echo ""
wait $PID1 $PID2 $PID3
echo "All training runs completed. Extracting Q-values..."
echo ""
# Extract initial Q-values from logs
echo "=== Run 1 - Initial Q-Values ==="
grep -E "Step 0.*Q-values:" /tmp/init_test_1.log | head -1 || echo "No Q-values found in Run 1"
echo ""
echo "=== Run 2 - Initial Q-Values ==="
grep -E "Step 0.*Q-values:" /tmp/init_test_2.log | head -1 || echo "No Q-values found in Run 2"
echo ""
echo "=== Run 3 - Initial Q-Values ==="
grep -E "Step 0.*Q-values:" /tmp/init_test_3.log | head -1 || echo "No Q-values found in Run 3"
echo ""
# Extract entropy seeds
echo "=== Entropy Seeds Used ==="
echo "Run 1:"
grep "Device RNG seeded with entropy:" /tmp/init_test_1.log | head -1 || echo "No seed found"
echo "Run 2:"
grep "Device RNG seeded with entropy:" /tmp/init_test_2.log | head -1 || echo "No seed found"
echo "Run 3:"
grep "Device RNG seeded with entropy:" /tmp/init_test_3.log | head -1 || echo "No seed found"
echo ""
echo "=== Validation ==="
echo "SUCCESS: If the Q-values and seeds are DIFFERENT across runs, the fix is working!"
echo "FAILURE: If the Q-values are IDENTICAL across runs, the issue persists."
echo ""
echo "Logs saved to: /tmp/init_test_{1,2,3}.log"