The 4-branch DQN (direction x magnitude) had 3 degenerate variants (Short25, Flat, Long25) that all mapped to 0.0 target exposure when direction=Flat, causing 82% Flat collapse. Collapse these into a single Flat variant, giving 7 levels (ShortSmall/Half/Full, Flat, LongSmall/Half/Full) and 63 total factored actions (7x3x3). - ExposureLevel enum: 9 variants -> 7 (add direction/magnitude/from_dir_mag) - FactoredAction: 81 -> 63 total actions, from_index/to_index updated - DQN epsilon-greedy: use from_dir_mag() instead of dir*3+mag indexing - DQN config: num_actions default 9 -> 7 - PPO action space: 45 -> 63 actions, action masking updated - Signal adapter CUDA kernel: 5-bin -> 7-bin exposure aggregation - All tests updated for new variant names and index ranges Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
384 lines
11 KiB
Rust
384 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 DQN action masking with position limits
|
|
//!
|
|
//! Tests verify that:
|
|
//! 1. Actions are correctly masked based on current position
|
|
//! 2. Position limits prevent invalid actions from being selected
|
|
//! 3. Masked action count tracking and logging works correctly
|
|
|
|
use ml::dqn::action_space::{get_valid_action_mask, ExposureLevel, FactoredAction};
|
|
|
|
#[test]
|
|
fn test_action_masking_at_positive_limit() {
|
|
// With max_position=1.0, only actions with |exposure| <= 1.0 are valid
|
|
let mask = get_valid_action_mask(2.0, 1.0);
|
|
|
|
// Long100 (exposure=+1.0) should be valid (exactly at limit)
|
|
for idx in 36..=44 {
|
|
assert!(
|
|
mask[idx],
|
|
"Long100 action {} should be valid (exposure=1.0 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long50 (exposure=+0.5) should be valid
|
|
for idx in 27..=35 {
|
|
assert!(
|
|
mask[idx],
|
|
"Long50 action {} should be valid (exposure=0.5 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Flat (exposure=0.0) should be valid
|
|
for idx in 18..=26 {
|
|
assert!(
|
|
mask[idx],
|
|
"Flat action {} should be valid (exposure=0.0 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Short actions (exposure=-0.5 to -1.0) should be valid
|
|
for idx in 0..=17 {
|
|
assert!(
|
|
mask[idx],
|
|
"Short action {} should be valid (|exposure| <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_action_masking_at_negative_limit() {
|
|
// With max_position=1.0, only actions with |exposure| <= 1.0 are valid
|
|
let mask = get_valid_action_mask(-2.0, 1.0);
|
|
|
|
// Short100 (exposure=-1.0) should be valid (exactly at limit)
|
|
for idx in 0..=8 {
|
|
assert!(
|
|
mask[idx],
|
|
"Short100 action {} should be valid (|exposure|=1.0 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Short50 (exposure=-0.5) should be valid
|
|
for idx in 9..=17 {
|
|
assert!(
|
|
mask[idx],
|
|
"Short50 action {} should be valid (|exposure|=0.5 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Flat (exposure=0.0) should be valid
|
|
for idx in 18..=26 {
|
|
assert!(
|
|
mask[idx],
|
|
"Flat action {} should be valid (exposure=0.0 <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long actions (exposure=+0.5 to +1.0) should be valid
|
|
for idx in 27..=44 {
|
|
assert!(
|
|
mask[idx],
|
|
"Long action {} should be valid (|exposure| <= max=1.0)",
|
|
idx
|
|
);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_action_masking_at_zero_position() {
|
|
// At position 0.0, all actions should be valid
|
|
let mask = get_valid_action_mask(0.0, 2.0);
|
|
|
|
for idx in 0..=44 {
|
|
assert!(mask[idx], "Action {} should be valid at position 0.0", idx);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_action_masking_prevents_long_at_max() {
|
|
// Position +1.5 with max 2.0:
|
|
// - Long100 (exposure=+1.0) would bring position to +1.0, which is valid
|
|
// - Long50 (exposure=+0.5) would bring position to +0.5, which is valid
|
|
// Note: Our implementation uses ABSOLUTE exposure, not deltas
|
|
let mask = get_valid_action_mask(1.5, 2.0);
|
|
|
|
// At position +1.5, actions with exposure <= 2.0 should be valid
|
|
// Since max exposure is ±1.0 (Long100/Short100), all actions are valid
|
|
for idx in 0..=44 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
let exposure = action.target_exposure();
|
|
if exposure.abs() <= 2.0 {
|
|
assert!(
|
|
mask[idx],
|
|
"Action {} (exposure={}) should be valid at position 1.5 with max 2.0",
|
|
idx, exposure
|
|
);
|
|
}
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_action_masking_prevents_short_at_min() {
|
|
// Position -1.5 with max 2.0:
|
|
// - Short100 (exposure=-1.0) would bring position to -1.0, which is valid
|
|
// - Short50 (exposure=-0.5) would bring position to -0.5, which is valid
|
|
let mask = get_valid_action_mask(-1.5, 2.0);
|
|
|
|
// At position -1.5, actions with |exposure| <= 2.0 should be valid
|
|
for idx in 0..=44 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
let exposure = action.target_exposure();
|
|
if exposure.abs() <= 2.0 {
|
|
assert!(
|
|
mask[idx],
|
|
"Action {} (exposure={}) should be valid at position -1.5 with max 2.0",
|
|
idx, exposure
|
|
);
|
|
}
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_flat_actions_always_valid() {
|
|
// Flat actions (exposure=0.0) should always be valid regardless of position
|
|
let positions = vec![-2.0, -1.5, -1.0, 0.0, 1.0, 1.5, 2.0];
|
|
|
|
for pos in positions {
|
|
let mask = get_valid_action_mask(pos, 2.0);
|
|
|
|
// Flat actions are indices 18-26
|
|
for idx in 18..=26 {
|
|
assert!(
|
|
mask[idx],
|
|
"Flat action {} should be valid at position {}",
|
|
idx, pos
|
|
);
|
|
}
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_masked_action_count() {
|
|
// With max_position=0.6, actions with |exposure| > 0.6 are masked
|
|
// Long100 (|exposure|=1.0) and Short100 (|exposure|=1.0) should be masked
|
|
let mask = get_valid_action_mask(0.0, 0.6);
|
|
let masked_count = mask.iter().filter(|&&v| !v).count();
|
|
|
|
// Long100 (9 actions: indices 36-44) + Short100 (9 actions: indices 0-8) = 18
|
|
assert_eq!(
|
|
masked_count, 18,
|
|
"Expected 18 masked actions with max_position=0.6"
|
|
);
|
|
|
|
// With max_position=1.0, all actions should be valid (max exposure is ±1.0)
|
|
let mask = get_valid_action_mask(0.0, 1.0);
|
|
let masked_count = mask.iter().filter(|&&v| !v).count();
|
|
|
|
assert_eq!(
|
|
masked_count, 0,
|
|
"Expected 0 masked actions with max_position=1.0"
|
|
);
|
|
|
|
// With max_position=2.0, all actions should be valid
|
|
let mask = get_valid_action_mask(0.0, 2.0);
|
|
let masked_count = mask.iter().filter(|&&v| !v).count();
|
|
|
|
assert_eq!(
|
|
masked_count, 0,
|
|
"Expected 0 masked actions with max_position=2.0"
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn test_restrictive_max_position() {
|
|
// With max_position=0.6, only Flat and ±50% exposure should be valid
|
|
let mask = get_valid_action_mask(0.0, 0.6);
|
|
|
|
// Short100 (exposure=-1.0) should be masked
|
|
for idx in 0..=8 {
|
|
assert!(
|
|
!mask[idx],
|
|
"Short100 action {} should be masked with max_position=0.6",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Short50 (exposure=-0.5) should be valid
|
|
for idx in 9..=17 {
|
|
assert!(
|
|
mask[idx],
|
|
"Short50 action {} should be valid with max_position=0.6",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Flat (exposure=0.0) should be valid
|
|
for idx in 18..=26 {
|
|
assert!(
|
|
mask[idx],
|
|
"Flat action {} should be valid with max_position=0.6",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long50 (exposure=+0.5) should be valid
|
|
for idx in 27..=35 {
|
|
assert!(
|
|
mask[idx],
|
|
"Long50 action {} should be valid with max_position=0.6",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long100 (exposure=+1.0) should be masked
|
|
for idx in 36..=44 {
|
|
assert!(
|
|
!mask[idx],
|
|
"Long100 action {} should be masked with max_position=0.6",
|
|
idx
|
|
);
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn test_action_mapping_correctness() {
|
|
// Verify that action indices map to correct exposure levels
|
|
// This ensures our masking logic uses correct indices
|
|
|
|
// Short100: indices 0-8
|
|
for idx in 0..=8 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
assert_eq!(
|
|
action.exposure,
|
|
ExposureLevel::ShortSmall,
|
|
"Action {} should have Short100 exposure",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Short50: indices 9-17
|
|
for idx in 9..=17 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
assert_eq!(
|
|
action.exposure,
|
|
ExposureLevel::ShortFull,
|
|
"Action {} should have Short50 exposure",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Flat: indices 18-26
|
|
for idx in 18..=26 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
assert_eq!(
|
|
action.exposure,
|
|
ExposureLevel::Flat,
|
|
"Action {} should have Flat exposure",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long50: indices 27-35
|
|
for idx in 27..=35 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
assert_eq!(
|
|
action.exposure,
|
|
ExposureLevel::LongSmall,
|
|
"Action {} should have Long50 exposure",
|
|
idx
|
|
);
|
|
}
|
|
|
|
// Long100: indices 36-44
|
|
for idx in 36..=44 {
|
|
let action = FactoredAction::from_index(idx).unwrap();
|
|
assert_eq!(
|
|
action.exposure,
|
|
ExposureLevel::LongFull,
|
|
"Action {} should have Long100 exposure",
|
|
idx
|
|
);
|
|
}
|
|
}
|