Files
foxhunt/crates/ml/tests/wave16q_ohlcv_mutation_test.rs
2026-04-04 10:02:37 +02:00

251 lines
8.5 KiB
Rust

#![allow(
clippy::assertions_on_constants,
clippy::assertions_on_result_states,
clippy::clone_on_copy,
clippy::decimal_literal_representation,
clippy::doc_markdown,
clippy::empty_line_after_doc_comments,
clippy::field_reassign_with_default,
clippy::get_unwrap,
clippy::identity_op,
clippy::inconsistent_digit_grouping,
clippy::indexing_slicing,
clippy::integer_division,
clippy::len_zero,
clippy::let_underscore_must_use,
clippy::manual_div_ceil,
clippy::manual_let_else,
clippy::manual_range_contains,
clippy::modulo_arithmetic,
clippy::needless_range_loop,
clippy::non_ascii_literal,
clippy::redundant_clone,
clippy::shadow_reuse,
clippy::shadow_same,
clippy::shadow_unrelated,
clippy::single_match_else,
clippy::str_to_string,
clippy::string_slice,
clippy::tests_outside_test_module,
clippy::too_many_lines,
clippy::unnecessary_wraps,
clippy::unseparated_literal_suffix,
clippy::use_debug,
clippy::useless_vec,
clippy::wildcard_enum_match_arm,
clippy::else_if_without_else,
clippy::expect_used,
clippy::missing_const_for_fn,
clippy::similar_names,
clippy::type_complexity,
clippy::collapsible_else_if,
clippy::doc_lazy_continuation,
clippy::items_after_test_module,
clippy::map_clone,
clippy::multiple_unsafe_ops_per_block,
clippy::unwrap_or_default,
clippy::assign_op_pattern,
clippy::needless_borrow,
clippy::println_empty_string,
clippy::unnecessary_cast,
clippy::used_underscore_binding,
clippy::create_dir,
clippy::implicit_saturating_sub,
clippy::exit,
clippy::expect_fun_call,
clippy::too_many_arguments,
clippy::unnecessary_map_or,
clippy::unwrap_used,
dead_code,
unused_imports,
unused_variables,
clippy::cloned_ref_to_slice_refs,
clippy::neg_multiply,
clippy::while_let_loop,
clippy::bool_assert_comparison,
clippy::excessive_precision,
clippy::trivially_copy_pass_by_ref,
clippy::op_ref,
clippy::redundant_closure,
clippy::unnecessary_lazy_evaluations,
clippy::if_then_some_else_none,
clippy::unnecessary_to_owned,
clippy::single_component_path_imports,
)]
// WAVE 16Q: OHLCV Mutation Bug Regression Test
//
// **Bug Description**: DQN training corrupts OHLCV bar close prices from realistic values
// ($4000-$5500 for ES futures) to preprocessed z-scores (0.001-2.0 range). This causes:
// 1. max_position explosion to 100M (should be 10-50)
// 2. P&L calculation completely broken
// 3. Reward signals meaningless
//
// **Root Cause**: target[2] (raw_current) receives preprocessed z-scores instead of raw dollars,
// suggesting either data aliasing, index mismatch, or hidden mutation in preprocessing/feature extraction.
//
// **Fix Strategy**: Extract raw close prices into separate vector before any preprocessing or
// feature extraction to guarantee data preservation.
use anyhow::Result;
use ml::trainers::DQNHyperparameters;
use ml::trainers::dqn::DQNTrainer;
use tracing::{info, warn};
#[tokio::test]
async fn test_ohlcv_raw_prices_preserved_after_preprocessing() -> Result<()> {
// Initialize trainer (preprocessing is always active)
let mut hyperparams = DQNHyperparameters::conservative();
hyperparams.preprocessing_window = 50;
hyperparams.preprocessing_clip_sigma = 3.0;
hyperparams.epochs = 1; // Single epoch test
let mut trainer = DQNTrainer::new(hyperparams)?;
// Use real DBN data (same as production)
let data_dir = "test_data/real/databento/ml_training";
// Run 1 epoch training to trigger preprocessing and feature extraction
// Use no-op checkpoint callback (just return empty string)
let noop_callback = |_epoch: usize, _data: Vec<u8>, _is_best: bool| -> Result<String> {
Ok(String::new())
};
let result = trainer.train(data_dir, "ES.FUT", noop_callback).await;
// Training may fail for other reasons, but we need to check OHLCV corruption
// even if training fails (the bug manifests during data loading)
if let Err(e) = result {
warn!(error = %e, "Training failed (may be expected)");
}
// Get validation data to inspect target values
let val_data = trainer.get_val_data();
// Validation: target[2] should contain raw prices ($4000-$5500), NOT z-scores (0.001-2.0)
let mut raw_prices_valid = 0;
let mut preprocessed_zscores = 0;
let mut min_raw = f64::MAX;
let mut max_raw = f64::MIN;
for (_features, target) in val_data.iter().take(100) {
let raw_current = target[2];
min_raw = min_raw.min(raw_current);
max_raw = max_raw.max(raw_current);
// ES futures trade around $4000-$5500
// Z-scores are in range -3 to +3 (clipped), typically 0.001 to 2.0
if raw_current.abs() > 100.0 {
// Realistic dollar price
raw_prices_valid += 1;
} else {
// Suspiciously small (likely z-score)
preprocessed_zscores += 1;
}
}
info!(
raw_prices_valid,
preprocessed_zscores,
min_raw = format_args!("{min_raw:.2}"),
max_raw = format_args!("{max_raw:.2}"),
"OHLCV validation results"
);
// ASSERTION: At least 90% of samples should have realistic raw prices
assert!(
raw_prices_valid >= 90,
"❌ BUG CONFIRMED: target[2] contains z-scores instead of raw prices! \
Valid: {}/100, Z-scores: {}/100, Range: [{:.2}, {:.2}]",
raw_prices_valid,
preprocessed_zscores,
min_raw,
max_raw
);
// ASSERTION: Min/max should be in realistic ES futures range ($4000-$5500)
assert!(
min_raw > 3000.0 && max_raw < 6000.0,
"❌ BUG CONFIRMED: Raw prices outside realistic range! \
Expected: $4000-$5500, Got: [{:.2}, {:.2}]",
min_raw,
max_raw
);
info!("OHLCV raw prices preserved correctly");
Ok(())
}
#[tokio::test]
async fn test_target_2_is_raw_price_not_zscore() -> Result<()> {
// Regression test: Verify target[2] contains raw dollars, not preprocessed z-scores
let mut hyperparams = DQNHyperparameters::conservative();
hyperparams.epochs = 1;
let mut trainer = DQNTrainer::new(hyperparams)?;
let data_dir = "test_data/real/databento/ml_training";
let noop_callback = |_: usize, _: Vec<u8>, _: bool| -> Result<String> { Ok(String::new()) };
let _ = trainer.train(data_dir, "ES.FUT", noop_callback).await;
let val_data = trainer.get_val_data();
assert!(!val_data.is_empty(), "Validation data should not be empty");
// Check first 10 samples
for (i, (_features, target)) in val_data.iter().take(10).enumerate() {
let raw_current = target[2];
info!(sample = i, target_2 = format_args!("{raw_current:.2}"), "sample target[2]");
// Assertion: raw_current should be in realistic price range, NOT z-score range
assert!(
raw_current.abs() > 100.0,
"Sample {}: target[2] = {:.6} looks like a z-score, not raw price",
i,
raw_current
);
}
info!("All samples have realistic raw prices in target[2]");
Ok(())
}
#[tokio::test]
async fn test_max_position_realistic_not_100m() -> Result<()> {
// Regression test: max_position should be 10-50, NOT 100M (indicates price corruption)
let mut hyperparams = DQNHyperparameters::conservative();
hyperparams.epochs = 1;
let mut trainer = DQNTrainer::new(hyperparams)?;
let data_dir = "test_data/real/databento/ml_training";
let noop_callback = |_: usize, _: Vec<u8>, _: bool| -> Result<String> { Ok(String::new()) };
let _ = trainer.train(data_dir, "ES.FUT", noop_callback).await;
let val_data = trainer.get_val_data();
// Simulate position calculation (simplified)
let mut max_position = 0.0f64;
for (_features, target) in val_data.iter() {
let price_change = (target[1] - target[0]).abs();
// Typical position sizing based on price change
let position = 10000.0 / price_change.max(0.01); // Avoid div by zero
max_position = max_position.max(position);
}
info!(max_position = format_args!("{max_position:.2}"), "Max position calculated");
// ASSERTION: max_position should be reasonable (10-50), NOT astronomical (100M)
assert!(
max_position < 1000.0,
"❌ BUG CONFIRMED: max_position = {:.2e} suggests price corruption (expected <1000)",
max_position
);
info!(max_position = format_args!("{max_position:.2}"), "max_position is realistic");
Ok(())
}