- Fixed all ML types to match actual backend response shapes:
MLStatus (logging state + decision files), MLDataStatus (datasets,
checkpoints, eval results), MLTrainStatus (phases, metrics, result),
PPOTrainStatus (metrics with win_rate/reward/losses), TournamentStatus
- Updated training API client with full parameter support for all endpoints
- Rewrote TrainingPage with 5 clear pipeline phases:
Phase 1: Data Collection (toggle logging, view decision files)
Phase 2: Dataset & Model Artifacts (splits, checkpoints, eval, model info)
Phase 3: Supervised Training (epochs/lr/batch/patience, live progress,
phase badges, real-time metrics, final eval results)
Phase 4: PPO Reinforcement Learning (full config: iterations, episodes,
clip epsilon, entropy coeff, opponent selection, live win rate)
Phase 5: Tournament Evaluation (episodes, playstyle, leaderboard table)
- All sections are collapsible with live phase badges and progress bars
- Advanced config toggles for supervised and PPO training
- Proper mutex: can't start PPO while supervised is running, etc.
- Auto-refresh model info and data status when training completes
- Redesigned Recommendations tab with EDHREC / From Collection sub-tabs
- EDHREC sub-tab: fetches average deck data, groups by role, shows owned/missing status
- Collection sub-tab: shows cards from collection grouped by type, shortfall indicators
- Both sub-tabs support: checkbox selection, bulk add to deck, single add buttons
- Role badges with color-coding (Ramp, Draw, Removal, BoardWipe, etc.)
- Card hover preview for EDHREC recommendations
- Show only owned filter for EDHREC results
- Fixed API client types: EdhRecsResponse, CollectionRecsResponse wrappers
- Added EdhRecCard and CollectionRecCard interfaces
Issue #1 — Winner's life total now captured:
- Parse verbose Forge log for life-total-change events
- Track last-known life per seat via 3 regex patterns
- Winner's final life set from log data (not left at default 40)
- Losers who died to mill/poison retain their actual life total
Issue #2 — Win condition classification formalized:
- New WinCondition enum (schema/WinCondition.java) with 8 categories
- Centralized Forge loss-reason-string → enum mapping (regex-based)
- UNKNOWN fallback with logging for unrecognized loss reasons
- All classification now goes through WinCondition.classify()
Issue #3 — JVM startup overhead reduced:
- Added -XX:+UseSerialGC (lighter GC for batch processes)
- Added -XX:TieredStopAtLevel=1 (faster JIT, less compilation)
- ForgeWorker stub class for future persistent batch mode
- Estimated reduction: ~2-3s → ~1-1.5s per game startup
Issue #4 — Single-threaded throughput optimization:
- New CLI flags: --ai-simplified, --ai-think-time, --benchmark
- JVM flags passed through to Forge subprocess
- Benchmark mode: run 1 game before batch, report wall/Forge/overhead times
- API support: aiSimplified, aiThinkTimeMs in StartRequest
Issue #5 — Threading, progress, and backpressure:
- --threads 0 or -1 auto-detects optimal count from CPU topology
- detectOptimalThreads() estimates physical cores (excludes HT)
- ProgressCallback interface for real-time sims/sec reporting
- Per-game progress emitted as [PROGRESS] lines, parsed by lab_api.py
- --max-queue flag for graceful backpressure (limits concurrent JVMs)
- Final throughput summary printed at batch completion
- StatusResponse now includes simsPerSec field
Also: .gitignore restored, lab_api.py updated with all new API fields
KoalaTrapLord