gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
d08ab461dbf2cccded54b620a046f7aed158b8c5
foxhunt/crates/ml/Cargo.toml
jgrusewski db874b1841 feat(foxhuntq): Phase 1c snapshot-resolution alpha + leakage fix + variable-dim fxcache 
Three things landing atomically because they're load-bearing for each other:

1. **Trend-scanning leakage fix** — trend_scanning.rs was emitting OLS slope+t-stat
   over a *forward* window [t, t+L]. With the Phase 1a label = sign(price[t+60]
   − price[t]), the forward feature window overlaps the label window, contaminating
   it. Purged walk-forward only sterilizes forward-looking *labels* that cross
   the train/val split, not forward-looking *features* that peek inside the same
   horizon the label measures. The leak inflated MLP accuracy from 0.49
   (legacy 74-dim baseline) to 0.75 — vanished to 0.50 after switching to a
   trailing window. Bounded the perfect-fit t-stat sentinel from ±1e6 → ±20
   (p<1e-30 is already meaningless); eliminated the 16k corruption-cap drops.

2. **Variable-dim alpha column** — fxcache schema now carries the alpha-feature
   width via metadata (`alpha_feature_dim`), not a compile-time constant. Same
   on-disk format hosts the 134-dim bar-level stack OR the 81-dim snapshot stack.
   Reader + auto-detect honor the metadata-declared dim; downstream MLP auto-sizes
   `in_dim`. Single schema, no forks.

3. **Snapshot pipeline (Phase 1c falsification)** — `snapshot_pipeline.rs`: 81-dim
   per-MBP10-snapshot extractor reusing 10 snapshot-native alpha blocks + 6 new
   snapshot-specific features (time-since-trade, time-since-snap, event-rate,
   spread-bps, L1-imbalance, microprice-mid drift). `precompute_features` gets
   `--row-unit snapshot` flag; emits one fxcache row per LOB update (1.97M rows
   from MBP-10 data vs 206K for bar mode).

**Smoke verdict on real data** (ES.FUT, 1.97M snapshots, 384K val):
- Bar-level honest alpha: accuracy=0.5005, AUC=0.5043 (no signal)
- **Snapshot-level alpha**: accuracy=0.5241, AUC=0.6849 (real signal, 384K val)
- GBM corroboration: accuracy=0.5401 (non-linear partitioning sees more)
- Horizon decay: alpha peaks at K=20-50 snapshots (~5-25ms), gone by K=500
- Regime-conditional: spread-Q4 quintile hits 0.752 accuracy on 76k samples

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-15 01:01:15 +02:00

293 lines
11 KiB
TOML
Raw Blame History

[package]
name = "ml"
version.workspace = true
edition.workspace = true
rust-version.workspace = true
authors.workspace = true
license.workspace = true
repository.workspace = true
homepage.workspace = true
documentation.workspace = true
publish.workspace = true
keywords.workspace = true
categories.workspace = true
[features]
# CUDA default: all ML training is GPU-only. Service crates on CPU nodes
# must opt out with `default-features = false, features = ["minimal-inference"]`.
#
# `default` includes `full-stack` so existing consumers depending on
# `ml.workspace = true` (no extra opts) continue to see the full module
# surface. Services that only need a subset should disable defaults and
# opt in to specific feature flags below (see services/*/Cargo.toml).
default = ["minimal-inference", "cuda", "full-stack"]
# PRODUCTION FEATURES - LIGHTWEIGHT ONLY
minimal-inference = [] # Minimal inference with no optional deps
financial = [] # Basic financial calculations
high-precision = ["rust_decimal/serde-float"]
# PERFORMANCE FEATURES - NO HEAVY ML
mimalloc-allocator = ["mimalloc"] # Fast memory allocator for 10-25% speedup
simd = [] # SIMD without heavy dependencies
# Storage and memory management features
gc = [] # Garbage collection features
s3-storage = ["ml-checkpoint/s3-storage", "aws-config", "aws-sdk-s3", "aws-types", "aws-credential-types", "urlencoding"] # S3 storage backend with AWS SDK
cuda = ["cudarc", "ml-core/cuda", "ml-dqn/cuda", "ml-ppo/cuda", "ml-supervised/cuda", "ml-ensemble/cuda", "ml-labeling/cuda", "ml-explainability/cuda", "ml-hyperopt/cuda"] # CUDA support — enabled by compile-training CI step via --features ml/cuda
nccl = ["cuda"] # NCCL multi-GPU data parallelism (requires NCCL library + cudarc nccl feature)
# ============================================================================
# OPT-IN MODULE FEATURES — gate optional ml-* sub-crates and the `pub mod`
# items in lib.rs that depend on them. Services should enable only what they
# actually `use`. Each leaf sub-crate maps to exactly one `pub mod` in lib.rs.
# ============================================================================
backtest-mod = ["dep:ml-backtesting"] # ml::backtesting
paper-trading-mod = ["dep:ml-paper-trading"] # ml::paper_trading
stress-testing-mod = ["dep:ml-stress-testing"] # ml::stress_testing
explainability-mod = ["dep:ml-explainability"] # ml::explainability
universe-mod = ["dep:ml-universe"] # ml::universe
regime-detection-mod = ["dep:ml-regime-detection"] # ml::regime_detection
validation-mod = ["dep:ml-validation"] # ml::validation
data-validation-mod = ["dep:ml-data-validation"] # ml::data_validation
# Convenience: full module surface (used by `default` to preserve existing
# behavior). Services that opt out of `default` and want everything can
# still enable just `full-stack`.
full-stack = [
"backtest-mod",
"paper-trading-mod",
"stress-testing-mod",
"explainability-mod",
"universe-mod",
"regime-detection-mod",
"validation-mod",
"data-validation-mod",
]
# ALL HEAVY ML FEATURES REMOVED:
# gpu, pytorch, linfa-ml - MOVED TO ml_training_service
# optimization, graph-models, reinforcement-learning - MOVED TO ml_training_service
# transformers-advanced - MOVED TO ml_training_service
[dependencies]
# Core async and utilities
tokio.workspace = true
futures.workspace = true
async-trait.workspace = true
clap.workspace = true # CLI argument parsing for train_tft binary
# Serialization and error handling
serde.workspace = true
serde_json.workspace = true
serde_yaml = "0.9" # YAML serialization for hyperparameter exports
toml.workspace = true
uuid.workspace = true
thiserror.workspace = true
anyhow.workspace = true
chrono.workspace = true
chrono-tz = "0.10" # Timezone support for market hours calculations (Wave C)
time = "0.3" # Required by dbn::TsSymbolMap for instrument_id → symbol resolution
csv = "1.3" # CSV serialization for action export (Wave 3 Task 3.1)
rand.workspace = true
rand_chacha = "0.3" # ChaCha RNG for reproducible Monte Carlo permutation tests
# System and I/O
memmap2.workspace = true
tempfile.workspace = true
tracing.workspace = true
tracing-subscriber.workspace = true # For train_tft binary logging
prometheus.workspace = true
reqwest.workspace = true
colored = "2.1" # Terminal color output for evaluation reports
# Internal workspace crates
# Always-on: required by inference, training_pipeline, safety, ensemble, and
# many cross-module references inside ml/src/.
ml-core.workspace = true
ml-ensemble.workspace = true
ml-dqn.workspace = true
ml-ppo.workspace = true
ml-supervised.workspace = true
ml-hyperopt.workspace = true
ml-features.workspace = true
ml-labeling.workspace = true
ml-regime.workspace = true
ml-checkpoint.workspace = true
ml-risk.workspace = true
ml-security.workspace = true
ml-asset-selection.workspace = true
ml-observability.workspace = true
# Opt-in via features (gated `pub mod` items in lib.rs). Services that don't
# use these modules don't pay the compile cost. See [features] above for
# which feature gates which crate / module.
ml-backtesting = { workspace = true, optional = true }
ml-paper-trading = { workspace = true, optional = true }
ml-stress-testing = { workspace = true, optional = true }
ml-explainability = { workspace = true, optional = true }
ml-universe = { workspace = true, optional = true }
ml-regime-detection = { workspace = true, optional = true }
ml-validation = { workspace = true, optional = true }
ml-data-validation = { workspace = true, optional = true }
config.workspace = true
common = { workspace = true, features = ["questdb"] }
risk = { path = "../risk" }
# Model loading functionality is in storage crate
storage = { path = "../storage" }
# Data crate for test helpers (dev-dependency in tests)
data = { path = "../data" }
# Database for model registry
sqlx.workspace = true
# candle-core, candle-nn, candle-optimisers — REMOVED (replaced by ml-core native CUDA autograd)
# HEAVY ML FRAMEWORKS REMOVED - MOVED TO ml_training_service
# ort (ONNX Runtime) - REMOVED (1000+ dependencies alone!)
# tch, torch-sys (PyTorch bindings) - REMOVED (500+ dependencies!)
# Mathematical libraries.
# ndarray's `blas` feature is intentionally not enabled — it requires
# libopenblas-dev on the build host, which the CI compile pool does not
# provide. Training hot paths run through CUDA cuBLAS on GPU, so the
# host-side BLAS is unnecessary.
ndarray = { workspace = true, features = ["rayon"] }
nalgebra = { version = "0.33", features = ["serde-serialize"] }
# MINIMAL statistics only - ALL HEAVY ML ALGORITHMS REMOVED
# linfa ecosystem (linfa, linfa-clustering, linfa-linear, linfa-reduction) - REMOVED (200+ deps)
# smartcore - REMOVED (100+ dependencies)
# Basic statistics - always included (not optional)
statrs.workspace = true # Required for statistical computations
rust_decimal.workspace = true
# gymnasium, rerun - REMOVED (RL frameworks moved to ml_training_service)
# cudarc — vendored fork (vendor/cudarc): has_async_alloc=false fix for cublasLtMatmul on H100
cudarc = { version = "0.19", optional = true, default-features = false, features = ["driver", "cublas", "cublaslt", "dynamic-linking", "std", "cuda-version-from-build-system", "f16"] }
rayon.workspace = true
crossbeam = { version = "0.8", features = ["std"] }
petgraph = { version = "0.6", features = ["serde"] } # Required for TGNN graphs
semver = "1.0"
lru.workspace = true # Required for model caching
# chronoutil, ta, polars - REMOVED or moved to workspace dependencies
# argmin, nlopt, ipopt - REMOVED (optimization frameworks moved to ml_training_service)
rand_distr.workspace = true
dbn.workspace = true # Databento Binary format for real market data loading
zstd.workspace = true # Zstd decompression for .dbn.zst trade files
databento = "0.34" # Databento API client for downloading data (includes async by default)
# Performance allocators
mimalloc = { version = "0.1", optional = true } # Fast memory allocator
dotenv = "0.15" # Load .env files for API keys
parking_lot = { version = "0.12", features = ["hardware-lock-elision"] }
dashmap = { workspace = true }
once_cell = "1.19"
lazy_static.workspace = true
flate2 = "1.0"
sha2 = "0.10"
safetensors = "0.7" # Direct dep for checkpoint metadata (config hash validation)
hmac = "0.12" # HMAC for checkpoint signatures (SEC-001 fix)
hex = "0.4" # Hex encoding for signatures
bincode = "1.3"
fastrand = "2.1"
# wide - REMOVED (SIMD moved to trading_engine)
num-traits = "0.2"
# Parquet I/O for feature caching (Wave 2 Agent 8)
# Updated to workspace version 56 to fix arrow-arith compilation conflict
parquet.workspace = true
arrow = { workspace = true, features = ["ipc"] } # ipc feature: fxcache uses Arrow IPC format
bytes = "1.5" # For Parquet in-memory serialization
num = "0.4"
libc = "0.2"
fs2 = "0.4"
num_cpus = "1.16"
approx.workspace = true
sysinfo = "0.33" # System information for benchmarks
# AWS SDK dependencies for S3 checkpoint storage (optional, s3-storage feature)
aws-config = { version = "1.1", optional = true }
aws-sdk-s3 = { version = "1.14", optional = true }
aws-types = { version = "1.1", optional = true }
aws-credential-types = { version = "1.1", optional = true }
urlencoding = { version = "2.1", optional = true }
# Bayesian optimization for hyperparameter tuning (using argmin instead of egobox due to ndarray conflict)
argmin = { version = "0.8", features = ["rayon"] } # Optimization framework with parallel execution
argmin-math = "0.3" # Math utilities for argmin
[dev-dependencies]
tokio-test = "0.4"
proptest = "1.5"
tempfile = "3.12"
futures-test = "0.3"
test-case = "3.0"
rstest = "0.22"
criterion = { version = "0.5", features = ["html_reports", "async_tokio"] }
fastrand = "2.1"
tokio = { workspace = true, features = ["test-util", "macros"] }
insta = "1.34" # Snapshot testing for ML outputs
serial_test = "3.0" # Sequential testing for GPU resources
tracing-subscriber = { version = "0.3", features = ["env-filter", "fmt"] }
rand_chacha = "0.3" # ChaCha RNG for hyperopt tests
[[example]]
name = "cuda_test"
path = "examples/cuda_test.rs"
[[example]]
name = "train_baseline_rl"
path = "examples/train_baseline_rl.rs"
[[example]]
name = "train_baseline_supervised"
path = "examples/train_baseline_supervised.rs"
[[example]]
name = "evaluate_baseline"
path = "examples/evaluate_baseline.rs"
[[example]]
name = "hyperopt_baseline_rl"
path = "examples/hyperopt_baseline_rl.rs"
[[example]]
name = "hyperopt_baseline_supervised"
path = "examples/hyperopt_baseline_supervised.rs"
[[test]]
name = "behavioral_suite"
path = "tests/behavioral/main.rs"
[[bench]]
name = "microstructure_bench"
harness = false
[[bench]]
name = "alternative_bars_bench"
harness = false
[[bench]]
name = "bench_feature_extraction"
harness = false
[lints]
workspace = true
Reference in New Issue View Git Blame Copy Permalink