- Reduce CI GPU test datasets 16x for walltime reduction - Reduce early-stop epochs 50→10, add --test-threads=1 - Serialize all GPU lib tests to prevent cuBLAS init race - Align state_dim to 16 for BF16 tensor core HMMA dispatch - BF16 precision tolerance in ml-dqn tests - Enable branching DQN + tracing subscriber in smoke tests - Prevent min_replay_size > buffer_size deadlock in early-stop tests - Prevent AutoReplaySizer from breaking gradient collapse warmup - Replace racy tokio::spawn checkpoint counter with AtomicUsize - Set warmup_steps=0 and max_training_steps_per_epoch=300 in early-stop tests - RealDataLoader respects TEST_DATA_DIR for CI PVC layout - Add collapse_warmup_capacity to gpu_smoketest DQNConfig - Drain CUDA context between test binaries - Detached HEAD checkout prevents local branch corruption - GPU pipeline tests: fix BF16 dtype and rank-1 squeeze assertions - OOD input handling tests use use_gpu: true Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
200 lines
6.3 KiB
Rust
200 lines
6.3 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,
|
|
)]
|
|
// Bug #20: Portfolio value normalization test
|
|
// Tests that portfolio value is normalized by initial_capital
|
|
// to prevent 100,000x feature imbalance
|
|
|
|
use anyhow::Result;
|
|
use ml::dqn::portfolio_tracker::PortfolioTracker;
|
|
use tracing::info;
|
|
|
|
#[test]
|
|
fn test_portfolio_value_normalized_to_baseline() -> Result<()> {
|
|
// Bug #20: Portfolio value should be normalized to ~1.0 baseline
|
|
let initial_capital = 100_000.0;
|
|
let avg_spread = 0.0001;
|
|
let cash_reserve_percent = 0.0;
|
|
|
|
let tracker = PortfolioTracker::new(initial_capital, avg_spread, cash_reserve_percent);
|
|
|
|
// At start, portfolio value = initial_capital
|
|
let features = tracker.get_portfolio_features(4000.0);
|
|
|
|
// Bug #20: Feature should be 1.0 (normalized), NOT 100,000
|
|
assert_eq!(features.len(), 3, "Should have 3 portfolio features");
|
|
|
|
let portfolio_feature = features[0];
|
|
|
|
// After fix, this should be ~1.0 (normalized by initial_capital)
|
|
// Before fix, this would be 100,000.0 (raw value)
|
|
assert!(
|
|
(portfolio_feature - 1.0).abs() < 0.01,
|
|
"Portfolio value should be normalized to 1.0, got: {}",
|
|
portfolio_feature
|
|
);
|
|
|
|
info!(portfolio_feature, "Portfolio value normalized correctly");
|
|
|
|
Ok(())
|
|
}
|
|
|
|
#[test]
|
|
fn test_all_portfolio_features_similar_scale() -> Result<()> {
|
|
// Bug #20: All portfolio features should be in similar scale
|
|
// No 100,000x imbalance
|
|
let initial_capital = 100_000.0;
|
|
let avg_spread = 0.0001;
|
|
let cash_reserve_percent = 0.0;
|
|
|
|
let tracker = PortfolioTracker::new(initial_capital, avg_spread, cash_reserve_percent);
|
|
|
|
let features = tracker.get_portfolio_features(4000.0);
|
|
|
|
// Check all features are in reasonable scale (NOT 100,000x difference)
|
|
for (i, &feature) in features.iter().enumerate() {
|
|
assert!(
|
|
feature.abs() < 10.0,
|
|
"Feature {} should be in reasonable scale, got: {}",
|
|
i,
|
|
feature
|
|
);
|
|
}
|
|
|
|
info!(?features, "All portfolio features in similar scale");
|
|
|
|
Ok(())
|
|
}
|
|
|
|
#[test]
|
|
fn test_feature_scale_consistency() -> Result<()> {
|
|
// Bug #20: Verify portfolio features don't dominate other features
|
|
// The key test is that portfolio value is NOT 100,000 (raw value)
|
|
let initial_capital = 100_000.0;
|
|
let avg_spread = 0.0001;
|
|
let cash_reserve_percent = 0.0;
|
|
|
|
let tracker = PortfolioTracker::new(initial_capital, avg_spread, cash_reserve_percent);
|
|
|
|
let features = tracker.get_portfolio_features(4000.0);
|
|
|
|
// The key fix: Portfolio value should be ~1.0 (normalized), NOT 100,000
|
|
let portfolio_value = features[0];
|
|
|
|
// Before Bug #20 fix: Would be 100,000.0 (raw value)
|
|
// After Bug #20 fix: Should be ~1.0 (normalized)
|
|
assert!(
|
|
portfolio_value.abs() < 10.0,
|
|
"Portfolio value should be normalized, got: {}",
|
|
portfolio_value
|
|
);
|
|
|
|
// Check that portfolio value doesn't dwarf other features by 100,000x
|
|
// (spread is intentionally small, but that's OK - key is portfolio isn't massive)
|
|
let max_feature = features.iter().copied().fold(f32::NEG_INFINITY, f32::max);
|
|
|
|
// Before fix: max_feature would be 100,000 (raw portfolio value)
|
|
// After fix: max_feature should be ~1.0 (normalized portfolio value)
|
|
assert!(
|
|
max_feature < 10.0,
|
|
"Max feature should be in normalized scale, got: {}",
|
|
max_feature
|
|
);
|
|
|
|
info!(max_feature, portfolio_value, "Feature scale check passed");
|
|
|
|
Ok(())
|
|
}
|
|
|
|
#[test]
|
|
fn test_portfolio_feature_format() -> Result<()> {
|
|
// Test that get_portfolio_features returns 3 features:
|
|
// [portfolio_value, position_normalized, spread]
|
|
let tracker = PortfolioTracker::new(100_000.0, 0.0001, 0.0);
|
|
let features = tracker.get_portfolio_features(4000.0);
|
|
|
|
assert_eq!(features.len(), 3, "Should return exactly 3 features");
|
|
|
|
// Feature 0: Portfolio value (should be normalized to ~1.0)
|
|
// Feature 1: Position (normalized by max_position)
|
|
// Feature 2: Spread
|
|
|
|
info!(?features, "Portfolio features");
|
|
info!(value = features[0], "Feature 0 (portfolio value)");
|
|
info!(position = features[1], "Feature 1 (position)");
|
|
info!(spread = features[2], "Feature 2 (spread)");
|
|
|
|
Ok(())
|
|
}
|