Files
foxhunt/crates/ml/tests/dqn_advanced_metrics_integration_test.rs
jgrusewski ca4c38d921 fix(tests): CI GPU test stability, walltime reduction, BF16 tolerance
- 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>
2026-03-15 12:00:13 +01:00

325 lines
11 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,
)]
//! Integration tests for Advanced Performance Metrics
//!
//! Validates:
//! 1. Sortino ratio calculated correctly (downside deviation focus)
//! 2. Calmar ratio calculated correctly (return / max drawdown)
//! 3. VaR (95%) calculated correctly (Value at Risk)
//! 4. CVaR (95%) calculated correctly (Conditional VaR / Expected Shortfall)
//! 5. Composite objective uses all 4 metrics
//!
//! Advanced Metrics:
//! - Sortino Ratio: return / downside_deviation (punishes downside volatility)
//! - Calmar Ratio: annualized_return / max_drawdown (risk-adjusted)
//! - VaR (95%): 5th percentile of returns (worst 5% threshold)
//! - CVaR (95%): avg of returns below VaR (tail risk)
use ml::MLError;
use tracing::info;
/// Test 1: Verify Sortino ratio calculation
#[test]
fn test_sortino_ratio_calculation() -> Result<(), MLError> {
// Sample returns: [+2%, -1%, +3%, -0.5%, +1%]
let returns = vec![0.02, -0.01, 0.03, -0.005, 0.01];
// Calculate mean return
let mean_return = returns.iter().sum::<f64>() / returns.len() as f64;
// Calculate downside deviation (only negative returns)
let downside_returns: Vec<f64> = returns
.iter()
.filter(|&&r| r < 0.0)
.map(|&r| r)
.collect();
let downside_deviation = if !downside_returns.is_empty() {
let mean_downside = downside_returns.iter().sum::<f64>() / downside_returns.len() as f64;
let variance: f64 = downside_returns
.iter()
.map(|&r| (r - mean_downside).powi(2))
.sum::<f64>()
/ downside_returns.len() as f64;
variance.sqrt()
} else {
0.0
};
let sortino_ratio = if downside_deviation > 0.0 {
mean_return / downside_deviation
} else {
f64::INFINITY // No downside = infinite Sortino
};
info!(mean_return, downside_deviation, sortino_ratio, "Sortino ratio calculation");
assert!(
mean_return > 0.0,
"Mean return should be positive for this sample"
);
assert!(
downside_deviation > 0.0,
"Downside deviation should be positive (2 negative returns)"
);
assert!(
sortino_ratio.is_finite() && sortino_ratio > 0.0,
"Sortino ratio should be finite and positive"
);
info!(sortino_ratio, "Sortino ratio calculated correctly");
Ok(())
}
/// Test 2: Verify Calmar ratio calculation
#[test]
fn test_calmar_ratio_calculation() -> Result<(), MLError> {
// Sample equity curve (cumulative returns)
let equity = vec![100.0, 102.0, 101.0, 104.0, 103.0, 108.0];
// Calculate annualized return (assume 252 trading days)
let initial_equity = equity[0];
let final_equity = equity[equity.len() - 1];
let total_return = (final_equity - initial_equity) / initial_equity;
let periods = equity.len() - 1;
let annualized_return = total_return * (252.0 / periods as f64);
// Calculate maximum drawdown
let mut max_equity = equity[0];
let mut max_drawdown = 0.0;
for &current_equity in &equity {
if current_equity > max_equity {
max_equity = current_equity;
}
let drawdown = (max_equity - current_equity) / max_equity;
if drawdown > max_drawdown {
max_drawdown = drawdown;
}
}
let calmar_ratio = if max_drawdown > 0.0 {
annualized_return / max_drawdown
} else {
f64::INFINITY // No drawdown = infinite Calmar
};
info!(annualized_return, max_drawdown, calmar_ratio, "Calmar ratio calculation");
assert!(
total_return > 0.0,
"Total return should be positive (100 → 108)"
);
assert!(
max_drawdown > 0.0,
"Max drawdown should be positive (drawdowns exist)"
);
assert!(
calmar_ratio.is_finite() && calmar_ratio > 0.0,
"Calmar ratio should be finite and positive"
);
info!(calmar_ratio, "Calmar ratio calculated correctly");
Ok(())
}
/// Test 3: Verify VaR (95%) calculation
#[test]
fn test_var_95_calculation() -> Result<(), MLError> {
// Sample returns (sorted for percentile calculation)
let mut returns = vec![
0.05, 0.03, 0.02, 0.01, 0.00, -0.01, -0.02, -0.03, -0.04, -0.05,
0.04, 0.02, 0.01, -0.01, -0.02, -0.03, 0.03, 0.01, -0.01, -0.06,
];
returns.sort_by(|a, b| a.partial_cmp(b).unwrap());
// VaR (95%): 5th percentile (worst 5% threshold)
let percentile_idx = (returns.len() as f64 * 0.05).ceil() as usize - 1;
let var_95 = returns[percentile_idx.min(returns.len() - 1)];
info!(first_5 = ?&returns[..5], percentile_idx, var_95, "VaR 95% calculation");
assert!(
var_95 < 0.0,
"VaR (95%) should be negative (represents worst 5% loss)"
);
info!(var_95, "VaR 95% calculated correctly");
Ok(())
}
/// Test 4: Verify CVaR (95%) calculation
#[test]
fn test_cvar_95_calculation() -> Result<(), MLError> {
// Sample returns (sorted for tail risk calculation)
let mut returns = vec![
0.05, 0.03, 0.02, 0.01, 0.00, -0.01, -0.02, -0.03, -0.04, -0.05,
0.04, 0.02, 0.01, -0.01, -0.02, -0.03, 0.03, 0.01, -0.01, -0.06,
];
returns.sort_by(|a, b| a.partial_cmp(b).unwrap());
// VaR (95%): 5th percentile
let percentile_idx = (returns.len() as f64 * 0.05).ceil() as usize - 1;
let var_95 = returns[percentile_idx.min(returns.len() - 1)];
// CVaR (95%): average of returns below VaR (expected shortfall)
let tail_returns: Vec<f64> = returns.iter().filter(|&&r| r <= var_95).copied().collect();
let cvar_95 = if !tail_returns.is_empty() {
tail_returns.iter().sum::<f64>() / tail_returns.len() as f64
} else {
var_95 // Fallback to VaR if no tail returns
};
info!(var_95, ?tail_returns, cvar_95, "CVaR 95% calculation");
assert!(
cvar_95 <= var_95,
"CVaR should be ≤ VaR (tail average ≤ tail threshold)"
);
info!(cvar_95, "CVaR 95% calculated correctly");
Ok(())
}
/// Test 5: Verify composite objective uses all 4 metrics
#[test]
fn test_composite_objective_all_metrics() -> Result<(), MLError> {
// Calculate all 4 metrics for a sample strategy
let returns = vec![0.02, -0.01, 0.03, -0.005, 0.01, 0.02, -0.02];
// 1. Sortino ratio
let mean_return = returns.iter().sum::<f64>() / returns.len() as f64;
let downside_returns: Vec<f64> = returns.iter().filter(|&&r| r < 0.0).copied().collect();
let downside_dev = if !downside_returns.is_empty() {
let mean_down = downside_returns.iter().sum::<f64>() / downside_returns.len() as f64;
let var: f64 = downside_returns
.iter()
.map(|&r| (r - mean_down).powi(2))
.sum::<f64>()
/ downside_returns.len() as f64;
var.sqrt()
} else {
1.0
};
let sortino = mean_return / downside_dev;
// 2. Calmar ratio (simplified)
let total_return = returns.iter().sum::<f64>() / returns.len() as f64;
let max_drawdown = 0.02; // Assume 2% max drawdown
let calmar = total_return / max_drawdown;
// 3. VaR (95%)
let mut sorted_returns = returns.clone();
sorted_returns.sort_by(|a, b| a.partial_cmp(b).unwrap());
let var_idx = (sorted_returns.len() as f64 * 0.05).ceil() as usize - 1;
let var_95 = sorted_returns[var_idx.min(sorted_returns.len() - 1)];
// 4. CVaR (95%)
let tail: Vec<f64> = sorted_returns.iter().filter(|&&r| r <= var_95).copied().collect();
let cvar_95 = tail.iter().sum::<f64>() / tail.len() as f64;
// Composite objective: weighted average
let weight_sortino = 0.3;
let weight_calmar = 0.3;
let weight_var = 0.2;
let weight_cvar = 0.2;
// Normalize metrics to [0, 1] range (assume Sortino/Calmar ~ 0-5, VaR/CVaR ~ -0.1 to 0)
let norm_sortino = (sortino / 5.0).clamp(0.0, 1.0);
let norm_calmar = (calmar / 5.0).clamp(0.0, 1.0);
let norm_var = (1.0 + var_95 / 0.1).clamp(0.0, 1.0); // Higher is better (less negative)
let norm_cvar = (1.0 + cvar_95 / 0.1).clamp(0.0, 1.0);
let composite_objective = weight_sortino * norm_sortino
+ weight_calmar * norm_calmar
+ weight_var * norm_var
+ weight_cvar * norm_cvar;
info!(
sortino, norm_sortino, calmar, norm_calmar,
var_95, norm_var, cvar_95, norm_cvar, composite_objective,
"Composite objective from all 4 metrics"
);
assert!(
composite_objective >= 0.0 && composite_objective <= 1.0,
"Composite objective should be in [0, 1], got {}",
composite_objective
);
info!(composite_objective, "Composite objective uses all 4 metrics");
Ok(())
}