gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(architectural): volume_bar_size in cache key + OFI front-month filter

Browse Source 
## Two architectural cleanups, both surfaced by the wgdc8 experiment

### Part 1: volume_bar_size in cache key

Mirrors the imbalance_bar_threshold/ewma_alpha fix from `f7718b376`. The
volume bar size constant (100 contracts/bar) was previously hardcoded and
not in the fxcache key. Tuning it would have hit the same fossilization
bug as imbalance_bar_threshold did pre-fix.

Changes:
- `Hyperparams.volume_bar_size: u64` field added (default 100, matches
  `DEFAULT_VOLUME_BAR_SIZE` for backwards compat).
- TrainingProfile loader reads `volume_bar_size` TOML key.
- `calculate_dbn_cache_key_full` signature 7 → 8 args. Hashed via
  `to_le_bytes()`. Test `test_cache_key_includes_volume_bar_size`
  added; passes alongside the 5 existing tests.
- 4 callers updated atomically (per `feedback_no_partial_refactor`):
  `discover_and_load`, `data_loading.rs:146`, `train_baseline_rl.rs:599`,
  `precompute_features.rs:259,720`.
- `data_loading.rs:279` now passes `self.hyperparams.volume_bar_size`
  to `build_volume_bars` instead of the hardcoded `DEFAULT_VOLUME_BAR_SIZE`.
- New `--volume-bar-size` CLI arg on both binaries (default 100).
- New Argo workflow params `volume-bar-size` (default "100") and
  `data-source` (default "mbp10") on both `train-template.yaml` and
  `train-multi-seed-template.yaml`. Threaded into precompute + trainer
  invocations.
- `scripts/argo-train.sh` exposes `--volume-bar-size <n>` and
  `--data-source <s>` for ad-hoc overrides.

### Part 2: OFI front-month filter (latent bug fix)

`crates/ml/examples/precompute_features.rs:539-557` (the OFI/VPIN/Kyle's
Lambda computation branch when MBP-10 + trades data is available) was
loading trades unfiltered for per-bar microstructure feature computation.
The volume bar formation path filters front-month per-file (line 354), but
the OFI path did not.

Effect pre-fix: during contract rollover windows (e.g., ESZ24 → ESH25),
OFI per-bar microstructure features included trades from BOTH contracts
simultaneously, distorting VPIN, Kyle's Lambda, and trade imbalance
signals. Severity in production: small (front-month dominates ES.FUT
volume by 10-100×) but real and present in every prior MBP-10+trades
production run.

Fix: mirror the per-file `filter_front_month` call from the volume bar
path. Volume bar formation and OFI computation now both see the same
in-month trade tape. Added log line shows raw vs filtered count per file
for transparency.

## Why bundled

Both fixes touch trade-data plumbing in `precompute_features.rs` and the
fxcache key contract. Per `feedback_no_partial_refactor`, related
architectural cleanups land atomically. Both surfaced from the same
wgdc8 audit; bundling avoids two cache-key-invalidating commits in
sequence (each would force full fxcache regen).

## Compatibility

- `volume_bar_size` defaults to 100 → existing wgdc7-equivalent runs
  reproduce, but with a *new* fxcache key (the f7718b376-era cache file
  is unreachable; harmless, can GC manually).
- OFI fix is strictly more correct; no opt-out needed. Existing models
  trained on contaminated OFI features may show slight feature
  distribution drift on first cache regen — expected, not a regression.
- `data_source = "ohlcv"` Argo param now possible; routes precompute
  through volume bar branch directly. wgdc8 experiment uses this to test
  bar resolution sensitivity at volume_bar_size=500 (5× DEFAULT).

Tests: 6/6 feature_cache tests pass. Workspace + examples compile clean.
Audit-doc: `docs/dqn-wire-up-audit.md` updated.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
jgrusewski
2026-05-09 22:55:33 +02:00
parent f7718b3761
commit abd7e533bc
11 changed files with 181 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
crates/ml/examples/precompute_features.rs
View File

@@ -161,6 +161,12 @@ struct Opts {
#[arg(long, default_value_t = 0.1)]
imbalance_bar_ewma_alpha: f64,
/// Volume bar size: contracts of one-sided volume per bar. Used when
/// `data_source != "mbp10"`. Default 100 matches `DEFAULT_VOLUME_BAR_SIZE`.
/// MUST match the trainer's value for cache HIT.
#[arg(long, default_value_t = 100)]
volume_bar_size: u64,
/// Skip confirmation prompt
#[arg(long)]
yes: bool,
@@ -278,6 +284,7 @@ async fn main() -> Result<()> {
&opts.data_source,
opts.imbalance_bar_threshold,
opts.imbalance_bar_ewma_alpha,
opts.volume_bar_size,
).context("Failed to compute cache key")?;
let early_check_path = output_dir.join(format!("{hex_key_early}.fxcache"));
if early_check_path.exists() {
@@ -387,9 +394,9 @@ async fn main() -> Result<()> {
);
bars
} else {
let bars = ml::features::build_volume_bars(&all_front_month_trades, ml::features::DEFAULT_VOLUME_BAR_SIZE);
let bars = ml::features::build_volume_bars(&all_front_month_trades, opts.volume_bar_size);
info!("Volume bars built: {} bars ({} contracts/bar) in {:.1}s",
bars.len(), ml::features::DEFAULT_VOLUME_BAR_SIZE, t1.elapsed().as_secs_f64());
bars.len(), opts.volume_bar_size, t1.elapsed().as_secs_f64());
bars
};
@@ -528,7 +535,17 @@ async fn main() -> Result<()> {
all_snapshots.sort_by_key(|s| s.timestamp);
info!("Loaded {} MBP-10 snapshots in {:.1}s", all_snapshots.len(), t2.elapsed().as_secs_f64());
// Load trades (parallel)
// Load trades (parallel) and filter front-month per-file.
//
// Pre-2026-05-09: this branch loaded trades unfiltered, leaking
// off-contract trade activity into the per-bar OFI/VPIN/Kyle's
// Lambda computations during contract rollover windows. Volume
// bar formation already filters front-month at line 354 of this
// file; the OFI side did not. Audit confirmed the latent bug
// affected every prior MBP-10+trades production run.
//
// Fix: mirror the per-file `filter_front_month` call from the
// volume bar path so OFI windows only see in-month trades.
let all_trades = if let Some(ref tdir) = trades_dir {
let mut trade_files = collect_dbn_files_recursive(tdir);
trade_files.sort();
@@ -536,7 +553,13 @@ async fn main() -> Result<()> {
use rayon::prelude::*;
trade_files.par_iter().filter_map(|path| {
match load_trades_sync(path) {
Ok(t) => { info!(" {} trades from {:?}", t.len(), path.file_name()); Some(t) }
Ok(t) => {
let raw_count = t.len();
let filtered = ml::features::filter_front_month(&t);
info!(" {} trades from {:?} ({} front-month)",
raw_count, path.file_name(), filtered.len());
Some(filtered)
}
Err(e) => { tracing::warn!(" Failed {:?}: {e}", path.file_name()); None }
}
}).collect()
@@ -771,6 +794,7 @@ async fn main() -> Result<()> {
&opts.data_source,
opts.imbalance_bar_threshold,
opts.imbalance_bar_ewma_alpha,
opts.volume_bar_size,
).context("Failed to compute cache key")?;
let cache_key: [u8; 32] = hex::decode(&hex_key)
.context("Invalid hex key")?

7
crates/ml/examples/train_baseline_rl.rs
View File

@@ -252,6 +252,12 @@ struct Args {
#[arg(long, default_value_t = 0.1)]
imbalance_bar_ewma_alpha: f64,
/// Volume bar size: contracts of one-sided volume per bar. Used when
/// `data_source != "mbp10"`. Default 100. MUST match precompute's value
/// for cache HIT.
#[arg(long, default_value_t = 100)]
volume_bar_size: u64,
/// Enable offline RL mode: train exclusively from a pre-collected dataset,
/// skipping online experience collection. Requires --dataset-path.
#[arg(long, default_value_t = false)]
@@ -619,6 +625,7 @@ fn run_training(args: &Args) -> Result<Vec<RlTrainingResult>> {
"mbp10",
args.imbalance_bar_threshold,
args.imbalance_bar_ewma_alpha,
args.volume_bar_size,
cache_dir_override.as_deref(),
);

53
crates/ml/src/feature_cache.rs
View File

@@ -15,23 +15,24 @@ use std::path::{Path, PathBuf};
/// (add / remove / modify) produces a different key.
///
/// Production defaults for the bar-formation params (matches `dqn-production.toml`):
/// `imbalance_bar_threshold = 0.5`, `imbalance_bar_ewma_alpha = 0.1`.
/// `imbalance_bar_threshold = 0.5`, `imbalance_bar_ewma_alpha = 0.1`,
/// `volume_bar_size = 100` (matches `DEFAULT_VOLUME_BAR_SIZE`).
pub fn calculate_dbn_cache_key(data_dir: &Path) -> Result<String> {
calculate_dbn_cache_key_full(data_dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1)
calculate_dbn_cache_key_full(data_dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100)
}
/// Extended cache key that includes MBP-10 and trades directories AND bar
/// formation parameters. The cache is invalidated when ANY data source
/// changes — OHLCV, MBP-10, trades — OR when bar-formation params change.
///
/// Bar-formation params (`imbalance_bar_threshold`, `imbalance_bar_ewma_alpha`)
/// are hashed in regardless of `data_source` because: (a) they uniquely identify
/// the bar stream the cache contains, (b) downstream features (e.g. log-returns,
/// targets) depend on bar boundaries, and (c) without this, two runs configured
/// with different thresholds will silently share the same cache file (canonical
/// 2026-05-09 audit: 14 prior production runs all collided on `a3f933aa...` /
/// `c07c960a...` keys regardless of TOML threshold value because the bar params
/// were not in the hash).
/// Bar-formation params (`imbalance_bar_threshold`, `imbalance_bar_ewma_alpha`,
/// `volume_bar_size`) are hashed in regardless of `data_source` because:
/// (a) they uniquely identify the bar stream the cache contains, (b) downstream
/// features (e.g. log-returns, targets) depend on bar boundaries, and
/// (c) without this, two runs configured with different params will silently
/// share the same cache file (canonical 2026-05-09 audit: 14 prior production
/// runs all collided on `a3f933aa...` / `c07c960a...` keys regardless of TOML
/// threshold value because the bar params were not in the hash).
pub fn calculate_dbn_cache_key_full(
data_dir: &Path,
mbp10_dir: Option<&Path>,
@@ -40,6 +41,7 @@ pub fn calculate_dbn_cache_key_full(
data_source: &str,
imbalance_bar_threshold: f64,
imbalance_bar_ewma_alpha: f64,
volume_bar_size: u64,
) -> Result<String> {
let mut hasher = Sha256::new();
@@ -54,6 +56,8 @@ pub fn calculate_dbn_cache_key_full(
hasher.update(b"|");
hasher.update(&imbalance_bar_ewma_alpha.to_le_bytes());
hasher.update(b"|");
hasher.update(&volume_bar_size.to_le_bytes());
hasher.update(b"|");
// Hash only file contents (size + mtime), not paths.
// This makes the key independent of CWD, relative/absolute paths,
@@ -101,8 +105,8 @@ mod tests {
fn test_cache_key_includes_symbol() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key_es = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key_nq = calculate_dbn_cache_key_full(dir, None, None, "NQ.FUT", "mbp10", 0.5, 0.1).unwrap();
let key_es = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
let key_nq = calculate_dbn_cache_key_full(dir, None, None, "NQ.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
assert_ne!(key_es, key_nq, "Different symbols must produce different cache keys");
}
@@ -110,8 +114,8 @@ mod tests {
fn test_cache_key_includes_data_source() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key_ohlcv = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "ohlcv", 0.5, 0.1).unwrap();
let key_mbp10 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key_ohlcv = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "ohlcv", 0.5, 0.1, 100).unwrap();
let key_mbp10 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
assert_ne!(key_ohlcv, key_mbp10, "Different data sources must produce different cache keys");
}
@@ -119,8 +123,8 @@ mod tests {
fn test_cache_key_includes_bar_threshold() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key_a = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key_b = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 2.5, 0.1).unwrap();
let key_a = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
let key_b = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 2.5, 0.1, 100).unwrap();
assert_ne!(key_a, key_b, "Different imbalance_bar_threshold must produce different cache keys");
}
@@ -128,17 +132,26 @@ mod tests {
fn test_cache_key_includes_bar_alpha() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key_a = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key_b = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 1.0).unwrap();
let key_a = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
let key_b = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 1.0, 100).unwrap();
assert_ne!(key_a, key_b, "Different imbalance_bar_ewma_alpha must produce different cache keys");
}
#[test]
fn test_cache_key_includes_volume_bar_size() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key_a = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "ohlcv", 0.5, 0.1, 100).unwrap();
let key_b = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "ohlcv", 0.5, 0.1, 500).unwrap();
assert_ne!(key_a, key_b, "Different volume_bar_size must produce different cache keys");
}
#[test]
fn test_cache_key_stable() {
let dir = Path::new("test_data/futures-baseline");
if !dir.exists() { return; }
let key1 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key2 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1).unwrap();
let key1 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
let key2 = calculate_dbn_cache_key_full(dir, None, None, "ES.FUT", "mbp10", 0.5, 0.1, 100).unwrap();
assert_eq!(key1, key2, "Same inputs must produce same key");
}
}

10
crates/ml/src/fxcache.rs
View File

@@ -690,8 +690,9 @@ pub fn norm_stats_path_for_key(
/// * `mbp10_dir` — Optional MBP-10 order book data directory
/// * `trades_dir` — Optional trades data directory
/// * `data_source` — Data source mode (`"mbp10"` or `"ohlcv"`)
/// * `imbalance_bar_threshold` — Bar formation threshold (must match producer's value)
/// * `imbalance_bar_ewma_alpha` — Bar formation EWMA alpha (must match producer's value)
/// * `imbalance_bar_threshold` — Imbalance bar formation threshold (must match producer's value)
/// * `imbalance_bar_ewma_alpha` — Imbalance bar EWMA alpha (must match producer's value)
/// * `volume_bar_size` — Volume bar contracts/bar (must match producer's value)
/// * `cache_dir_override` — Explicit cache directory (from CLI `--feature-cache-dir`)
pub fn discover_and_load(
data_dir: &Path,
@@ -701,6 +702,7 @@ pub fn discover_and_load(
data_source: &str,
imbalance_bar_threshold: f64,
imbalance_bar_ewma_alpha: f64,
volume_bar_size: u64,
cache_dir_override: Option<&Path>,
) -> Option<FxCacheData> {
// 1. Resolve cache directory
@@ -709,7 +711,7 @@ pub fn discover_and_load(
// 2. Compute cache key (includes symbol + data_source + bar params)
let key_hex = crate::feature_cache::calculate_dbn_cache_key_full(
data_dir, mbp10_dir, trades_dir, symbol, data_source,
imbalance_bar_threshold, imbalance_bar_ewma_alpha,
imbalance_bar_threshold, imbalance_bar_ewma_alpha, volume_bar_size,
)
.ok()?;
let key: [u8; 32] = hex::decode(&key_hex).ok()?.try_into().ok()?;
@@ -749,6 +751,7 @@ mod discover_tests {
"mbp10",
0.5,
0.1,
100,
None,
);
assert!(result.is_none());
@@ -780,6 +783,7 @@ mod discover_tests {
"mbp10",
0.5,
0.1,
100,
Some(&cache_dir),
);
// Strict match only — no "most recent" fallback

8
crates/ml/src/trainers/dqn/config.rs
View File

@@ -994,6 +994,13 @@ pub struct DQNHyperparameters {
/// Only used when `data_source = "mbp10"`.
pub imbalance_bar_ewma_alpha: f32,
/// Volume bar size: contracts of one-sided volume per bar. Higher =
/// fewer/longer bars. Used when `data_source != "mbp10"`. Default: 100
/// (matches `DEFAULT_VOLUME_BAR_SIZE` for backwards compatibility with
/// pre-2026-05-09 caches). Must match between `precompute_features` and
/// trainer for fxcache HIT (cache key includes this value).
pub volume_bar_size: u64,
// Offline RL mode (CQL/IQL on fixed datasets)
/// Offline RL mode: skip online experience collection, train from a fixed dataset.
/// Requires `dataset_path` to point to a pre-collected experience dataset.
@@ -1587,6 +1594,7 @@ impl DQNHyperparameters {
max_bars: 0, // 0 = unlimited
imbalance_bar_threshold: 100.0,
imbalance_bar_ewma_alpha: 0.1,
volume_bar_size: 100,
// Offline RL: disabled by default (online training with experience collection)
offline_mode: false,

8
crates/ml/src/trainers/dqn/data_loading.rs
View File

@@ -151,6 +151,7 @@ impl DQNTrainer {
&self.hyperparams.data_source,
f64::from(self.hyperparams.imbalance_bar_threshold),
f64::from(self.hyperparams.imbalance_bar_ewma_alpha),
self.hyperparams.volume_bar_size,
cache_dir_override,
) {
// OFI is unconditional — the model requires order flow data.
@@ -276,10 +277,11 @@ impl DQNTrainer {
all_front_month.sort_by_key(|t| t.timestamp);
let bars = crate::features::build_volume_bars(
&all_front_month, crate::features::DEFAULT_VOLUME_BAR_SIZE,
&all_front_month, self.hyperparams.volume_bar_size,
);
info!("Volume bars: {} bars from {} trades ({} front-month)",
bars.len(), total_raw, all_front_month.len());
info!("Volume bars: {} bars from {} trades ({} front-month, size={})",
bars.len(), total_raw, all_front_month.len(),
self.hyperparams.volume_bar_size);
if bars.is_empty() {
return Err(anyhow::anyhow!(

7
crates/ml/src/training_profile.rs
View File

@@ -95,6 +95,10 @@ pub struct TrainingSection {
/// EWMA alpha for adaptive imbalance threshold. Default: 0.1.
pub imbalance_bar_ewma_alpha: Option<f64>,
/// Volume bar size (contracts/bar). Used when `data_source != "mbp10"`.
/// Default: 100. Must match between precompute and trainer for cache HIT.
pub volume_bar_size: Option<u64>,
/// Maximum bars to load for training. 0 = unlimited (load all data).
/// Useful for fast smoketests on large datasets.
pub max_bars: Option<usize>,
@@ -811,6 +815,9 @@ impl DqnTrainingProfile {
if let Some(v) = t.imbalance_bar_ewma_alpha {
hp.imbalance_bar_ewma_alpha = v as f32;
}
if let Some(v) = t.volume_bar_size {
hp.volume_bar_size = v;
}
}
// [exploration]

50
docs/dqn-wire-up-audit.md
View File

@@ -2,6 +2,56 @@
**Status:** Populated during Plan 1 Task 6 (A.5 orphan audit). Updated on every commit per Invariant 7.
## 2026-05-09 — volume_bar_size in cache key + OFI front-month filter fix
`crates/ml/src/trainers/dqn/data_loading.rs` updated in two places:
1. Line 146: `discover_and_load` call now passes
`self.hyperparams.volume_bar_size` as the new bar-formation cache-key param
(8 → 9 args).
2. Line 279: `build_volume_bars` now uses `self.hyperparams.volume_bar_size`
instead of the `DEFAULT_VOLUME_BAR_SIZE` constant. Tuning this knob via
TOML now actually changes bar density (was previously a no-op via the
same fossilization pattern as `imbalance_bar_threshold` pre-fix).
### New CLI / TOML / Argo plumbing
- `Hyperparams.volume_bar_size: u64` field added (default 100, matches
`DEFAULT_VOLUME_BAR_SIZE`). TrainingProfile loader reads `volume_bar_size`
TOML key.
- `--volume-bar-size` CLI arg on `precompute_features` and
`train_baseline_rl` (default 100). Wired into Argo workflow params
`volume-bar-size` (default `"100"`) on both `train-template.yaml` and
`train-multi-seed-template.yaml`. `scripts/argo-train.sh` exposes
`--volume-bar-size <n>` for ad-hoc overrides.
- New Argo workflow param `data-source` (default `"mbp10"`) threaded
through to `precompute_features --data-source`. Exposed as
`--data-source <s>` on `argo-train.sh`. Lets experiments toggle volume
vs imbalance bar mode without TOML edits.
### Cache key
`calculate_dbn_cache_key_full` signature: 7 args → 8 args (added
`volume_bar_size: u64`). Hashed via `to_le_bytes()` after the imbalance
params. New unit test `test_cache_key_includes_volume_bar_size` pins the
contract; passes alongside the 5 existing tests.
### Latent OFI bug fix (atomic with this commit)
`crates/ml/examples/precompute_features.rs:539-557` (the OFI/VPIN/Kyle's
Lambda computation branch) was loading trades unfiltered for per-bar
microstructure feature computation. Pre-fix: the volume bar formation
path filtered front-month per-file (line 354), but the OFI path did not,
leaking off-contract trade activity into per-bar microstructure features
during contract rollover windows.
Fix: mirror the per-file `filter_front_month` call from the volume bar
path. Volume bar formation and OFI computation now both see the same
in-month trade tape. Severity in production: small (front-month dominates
ES.FUT trade volume by 10-100×, contamination ≪ 1%) but real and present
in every prior MBP-10+trades production run. Filed alongside the
volume_bar_size cache-key fix because both are architectural cleanups
the wgdc8 experiment surfaced.
## 2026-05-09 — fxcache key includes bar formation params + precompute_features actually USES data_source
`crates/ml/src/trainers/dqn/data_loading.rs:146` — `discover_and_load`

16
infra/k8s/argo/train-multi-seed-template.yaml
View File

@@ -100,6 +100,15 @@ spec:
value: "0.5"
- name: imbalance-bar-ewma-alpha
value: "0.1"
# Volume bar size (contracts/bar). Used when data-source != "mbp10".
# Default 100 matches DEFAULT_VOLUME_BAR_SIZE. Cache key includes this.
- name: volume-bar-size
value: "100"
# Data source mode: "mbp10" (imbalance bars from MBP-10) or "ohlcv"
# (volume bars from trades). Threaded into both precompute_features and
# the trainer so cache keys align.
- name: data-source
value: "mbp10"
volumes:
- name: git-ssh-key
@@ -353,8 +362,10 @@ spec:
--trades-data-dir /data/futures-baseline-trades \
--output-dir /feature-cache \
--symbol {{workflow.parameters.symbol}} \
--data-source {{workflow.parameters.data-source}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}} \
--yes; then
echo "=== Feature cache ready ==="
else
@@ -370,8 +381,10 @@ spec:
--trades-data-dir /data/futures-baseline-trades \
--output-dir /feature-cache \
--symbol {{workflow.parameters.symbol}} \
--data-source {{workflow.parameters.data-source}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}} \
--yes
fi
@@ -501,7 +514,8 @@ spec:
--seed "$SEED" \
--max-folds {{workflow.parameters.folds}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}}
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}}
echo "=== Training complete: seed=$SEED folds={{workflow.parameters.folds}} ==="

9
infra/k8s/argo/train-template.yaml
View File

@@ -73,6 +73,10 @@ spec:
value: "0.5"
- name: imbalance-bar-ewma-alpha
value: "0.1"
- name: volume-bar-size
value: "100"
- name: data-source
value: "mbp10"
volumes:
- name: git-ssh-key
@@ -373,8 +377,10 @@ spec:
--trades-data-dir /data/futures-baseline-trades \
--output-dir /feature-cache \
--symbol {{workflow.parameters.symbol}} \
--data-source {{workflow.parameters.data-source}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}} \
--yes; then
echo "=== Feature cache ready ==="
else
@@ -388,8 +394,10 @@ spec:
--trades-data-dir /data/futures-baseline-trades \
--output-dir /feature-cache \
--symbol {{workflow.parameters.symbol}} \
--data-source {{workflow.parameters.data-source}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}} \
--yes
echo "=== Feature cache regenerated ==="
fi
@@ -616,6 +624,7 @@ spec:
--epochs {{workflow.parameters.train-epochs}} \
--imbalance-bar-threshold {{workflow.parameters.imbalance-bar-threshold}} \
--imbalance-bar-ewma-alpha {{workflow.parameters.imbalance-bar-ewma-alpha}} \
--volume-bar-size {{workflow.parameters.volume-bar-size}} \
$HYPEROPT_FLAG
echo "=== Training complete ==="

12
scripts/argo-train.sh
View File

@@ -33,6 +33,8 @@ TAG=""
PROFILE=false
IMBALANCE_BAR_THRESHOLD=""
IMBALANCE_BAR_EWMA_ALPHA=""
VOLUME_BAR_SIZE=""
DATA_SOURCE=""
usage() {
cat <<EOF
@@ -61,6 +63,10 @@ Options:
--imbalance-bar-ewma-alpha <f> Override imbalance bar EWMA alpha (default: 0.1).
α=1.0 disables adaptation (this codebase's reversed
convention). MUST match between precompute and trainer.
--volume-bar-size <n> Override volume bar size in contracts (default: 100).
Used when --data-source != "mbp10". Cache key includes.
--data-source <s> Data source mode: "mbp10" or "ohlcv" (default: "mbp10").
Threaded into both precompute and trainer.
--profile Wrap training under nsys (NVIDIA Nsight Systems) and
upload .nsys-rep artefacts to MinIO bucket
foxhunt-training-artifacts/profiles/<sha>/. Forces the
@@ -100,6 +106,8 @@ while [[ $# -gt 0 ]]; do
--profile) PROFILE=true; shift ;;
--imbalance-bar-threshold) IMBALANCE_BAR_THRESHOLD="$2"; shift 2 ;;
--imbalance-bar-ewma-alpha) IMBALANCE_BAR_EWMA_ALPHA="$2"; shift 2 ;;
--volume-bar-size) VOLUME_BAR_SIZE="$2"; shift 2 ;;
--data-source) DATA_SOURCE="$2"; shift 2 ;;
--dry-run) DRY_RUN=true; shift ;;
-h|--help) usage ;;
*) echo "Unknown option: $1"; usage ;;
@@ -158,6 +166,8 @@ if [[ "$USE_MULTI_SEED" == "false" ]]; then
[[ "$SANITIZER" != "none" ]] && CMD="$CMD -p sanitizer=$SANITIZER"
[[ -n "$IMBALANCE_BAR_THRESHOLD" ]] && CMD="$CMD -p imbalance-bar-threshold=$IMBALANCE_BAR_THRESHOLD"
[[ -n "$IMBALANCE_BAR_EWMA_ALPHA" ]] && CMD="$CMD -p imbalance-bar-ewma-alpha=$IMBALANCE_BAR_EWMA_ALPHA"
[[ -n "$VOLUME_BAR_SIZE" ]] && CMD="$CMD -p volume-bar-size=$VOLUME_BAR_SIZE"
[[ -n "$DATA_SOURCE" ]] && CMD="$CMD -p data-source=$DATA_SOURCE"
$BASELINE && CMD="$CMD -p hyperopt-trials=0"
$WATCH && CMD="$CMD --watch"
@@ -291,6 +301,8 @@ CMD="$CMD -p profile=$PROFILE"
[[ -n "$TAG" ]] && CMD="$CMD --labels foxhunt-tag=$TAG"
[[ -n "$IMBALANCE_BAR_THRESHOLD" ]] && CMD="$CMD -p imbalance-bar-threshold=$IMBALANCE_BAR_THRESHOLD"
[[ -n "$IMBALANCE_BAR_EWMA_ALPHA" ]] && CMD="$CMD -p imbalance-bar-ewma-alpha=$IMBALANCE_BAR_EWMA_ALPHA"
[[ -n "$VOLUME_BAR_SIZE" ]] && CMD="$CMD -p volume-bar-size=$VOLUME_BAR_SIZE"
[[ -n "$DATA_SOURCE" ]] && CMD="$CMD -p data-source=$DATA_SOURCE"
[[ "$SANITIZER" != "none" ]] && CMD="$CMD -p sanitizer=$SANITIZER"
$BASELINE && CMD="$CMD -p hyperopt-trials=0"