49ad0050aa
BREAKING: Removes 746,569 lines of outdated documentation from root folder ## Summary - Deleted 2,060 report/documentation files from root folder - Kept only essential files: README.md, CLAUDE.md - Updated .gitignore and config/tarpaulin.toml - Reorganized config files into config/ directory ## Removed Content Categories - Agent reports (AGENT_*.md, AGENT*.txt) - Wave reports (WAVE_*.md, DQN_*.md) - Implementation summaries - Quick references and summaries - Test reports and validation docs - Deployment scripts (obsolete .sh files) - Legacy config files and logs ## Preserved - README.md - Main project documentation - CLAUDE.md - Claude Code configuration - docs/archive/ - Historical files for reference - docs/ folder - Current documentation - All source code unchanged 🐝 Hive Mind Collective Intelligence Cleanup 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
439 lines
15 KiB
Bash
Executable File
439 lines
15 KiB
Bash
Executable File
#!/usr/bin/env bash
|
||
#
|
||
# DQN Replay Pipeline Integration Test Script
|
||
#
|
||
# Purpose: Run comprehensive end-to-end tests for DQN evaluation pipeline
|
||
# Usage:
|
||
# ./test_dqn_replay_pipeline.sh # Run all tests (default)
|
||
# ./test_dqn_replay_pipeline.sh --quick # Run quick validation only
|
||
# ./test_dqn_replay_pipeline.sh --verbose # Enable verbose logging
|
||
# ./test_dqn_replay_pipeline.sh --ci # CI/CD mode (no ANSI colors)
|
||
#
|
||
# Exit codes:
|
||
# 0 - All tests passed
|
||
# 1 - Tests failed
|
||
# 2 - Setup error (missing dependencies, data files, etc.)
|
||
#
|
||
# Requirements:
|
||
# - Rust toolchain (cargo)
|
||
# - Test data: test_data/ES_FUT_unseen.parquet
|
||
# - Optional: CUDA for GPU testing
|
||
#
|
||
# Architecture:
|
||
# 1. Pre-flight checks (dependencies, test data)
|
||
# 2. Run integration tests (5 test suites)
|
||
# 3. Generate test report (summary, timing, failures)
|
||
# 4. Cleanup (optional)
|
||
|
||
set -euo pipefail
|
||
|
||
# ============================================================================
|
||
# Configuration
|
||
# ============================================================================
|
||
|
||
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||
PROJECT_ROOT="${SCRIPT_DIR}"
|
||
TEST_DATA_DIR="${PROJECT_ROOT}/test_data"
|
||
REQUIRED_TEST_FILE="ES_FUT_unseen.parquet"
|
||
|
||
# Default options
|
||
RUN_MODE="full"
|
||
VERBOSE=false
|
||
CI_MODE=false
|
||
CLEANUP=true
|
||
|
||
# ANSI color codes (disabled in CI mode)
|
||
GREEN='\033[0;32m'
|
||
RED='\033[0;31m'
|
||
YELLOW='\033[1;33m'
|
||
BLUE='\033[0;34m'
|
||
CYAN='\033[0;36m'
|
||
BOLD='\033[1m'
|
||
NC='\033[0m' # No Color
|
||
|
||
# ============================================================================
|
||
# Functions
|
||
# ============================================================================
|
||
|
||
print_banner() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "${CYAN}╔══════════════════════════════════════════════════════════════════════╗${NC}"
|
||
echo -e "${CYAN}║ DQN Replay Pipeline Integration Test Suite ║${NC}"
|
||
echo -e "${CYAN}║ Version: 1.0.0 ║${NC}"
|
||
echo -e "${CYAN}╚══════════════════════════════════════════════════════════════════════╝${NC}"
|
||
else
|
||
echo "==================================================================="
|
||
echo " DQN Replay Pipeline Integration Test Suite"
|
||
echo " Version: 1.0.0"
|
||
echo "==================================================================="
|
||
fi
|
||
echo ""
|
||
}
|
||
|
||
log_info() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "${BLUE}ℹ${NC} $1"
|
||
else
|
||
echo "[INFO] $1"
|
||
fi
|
||
}
|
||
|
||
log_success() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "${GREEN}✅${NC} $1"
|
||
else
|
||
echo "[PASS] $1"
|
||
fi
|
||
}
|
||
|
||
log_error() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "${RED}❌${NC} $1" >&2
|
||
else
|
||
echo "[FAIL] $1" >&2
|
||
fi
|
||
}
|
||
|
||
log_warn() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "${YELLOW}⚠️${NC} $1"
|
||
else
|
||
echo "[WARN] $1"
|
||
fi
|
||
}
|
||
|
||
log_step() {
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo -e "\n${BOLD}${CYAN}═══ $1 ═══${NC}\n"
|
||
else
|
||
echo ""
|
||
echo "=== $1 ==="
|
||
echo ""
|
||
fi
|
||
}
|
||
|
||
# ============================================================================
|
||
# Pre-flight Checks
|
||
# ============================================================================
|
||
|
||
check_dependencies() {
|
||
log_step "Pre-flight Checks"
|
||
|
||
# Check cargo
|
||
if ! command -v cargo &> /dev/null; then
|
||
log_error "cargo not found. Please install Rust toolchain."
|
||
exit 2
|
||
fi
|
||
log_success "cargo found: $(cargo --version)"
|
||
|
||
# Check CUDA availability (optional)
|
||
if command -v nvidia-smi &> /dev/null; then
|
||
log_success "CUDA available: $(nvidia-smi --query-gpu=name --format=csv,noheader | head -n1)"
|
||
else
|
||
log_warn "CUDA not available, tests will run on CPU"
|
||
fi
|
||
|
||
# Check test data directory
|
||
if [[ ! -d "${TEST_DATA_DIR}" ]]; then
|
||
log_error "Test data directory not found: ${TEST_DATA_DIR}"
|
||
exit 2
|
||
fi
|
||
log_success "Test data directory found: ${TEST_DATA_DIR}"
|
||
|
||
# Check for required test file
|
||
local test_file="${TEST_DATA_DIR}/${REQUIRED_TEST_FILE}"
|
||
if [[ ! -f "${test_file}" ]]; then
|
||
log_warn "Required test file not found: ${test_file}"
|
||
log_warn "Some tests may be skipped"
|
||
log_warn "Suggestion: Run data export script to generate test_data/ES_FUT_unseen.parquet"
|
||
else
|
||
local file_size=$(stat -f%z "${test_file}" 2>/dev/null || stat -c%s "${test_file}" 2>/dev/null || echo "unknown")
|
||
log_success "Test file found: ${test_file} (${file_size} bytes)"
|
||
fi
|
||
|
||
# Check disk space (need at least 500MB for temp files)
|
||
local available_space=$(df -m "${PROJECT_ROOT}" | tail -1 | awk '{print $4}')
|
||
if [[ "${available_space}" -lt 500 ]]; then
|
||
log_warn "Low disk space: ${available_space} MB available (recommended: 500 MB)"
|
||
else
|
||
log_success "Disk space: ${available_space} MB available"
|
||
fi
|
||
}
|
||
|
||
# ============================================================================
|
||
# Test Execution
|
||
# ============================================================================
|
||
|
||
run_integration_tests() {
|
||
log_step "Running Integration Tests"
|
||
|
||
local test_args=""
|
||
if [[ "${VERBOSE}" == "true" ]]; then
|
||
test_args="-- --nocapture --test-threads=1"
|
||
fi
|
||
|
||
local start_time=$(date +%s)
|
||
local test_results=()
|
||
local test_count=0
|
||
local pass_count=0
|
||
|
||
# Test 1: Full Pipeline
|
||
log_info "Test 1/5: Full Pipeline (export → load → backtest → validate)..."
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_full_replay_pipeline --release ${test_args}; then
|
||
log_success "Test 1: Full Pipeline PASSED"
|
||
test_results+=("PASS")
|
||
((pass_count++))
|
||
else
|
||
log_error "Test 1: Full Pipeline FAILED"
|
||
test_results+=("FAIL")
|
||
fi
|
||
((test_count++))
|
||
|
||
# Test 2: Timestamp Alignment
|
||
log_info "Test 2/5: Timestamp Alignment (>90% match rate)..."
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_timestamp_alignment --release ${test_args}; then
|
||
log_success "Test 2: Timestamp Alignment PASSED"
|
||
test_results+=("PASS")
|
||
((pass_count++))
|
||
else
|
||
log_error "Test 2: Timestamp Alignment FAILED"
|
||
test_results+=("FAIL")
|
||
fi
|
||
((test_count++))
|
||
|
||
# Test 3: Performance
|
||
log_info "Test 3/5: Performance (<30s constraint)..."
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_replay_performance --release ${test_args}; then
|
||
log_success "Test 3: Performance PASSED"
|
||
test_results+=("PASS")
|
||
((pass_count++))
|
||
else
|
||
log_error "Test 3: Performance FAILED"
|
||
test_results+=("FAIL")
|
||
fi
|
||
((test_count++))
|
||
|
||
# Test 4: Edge Cases
|
||
log_info "Test 4/5: Edge Cases (empty data, corrupt checkpoints, NaN/Inf)..."
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_replay_edge_cases --release ${test_args}; then
|
||
log_success "Test 4: Edge Cases PASSED"
|
||
test_results+=("PASS")
|
||
((pass_count++))
|
||
else
|
||
log_error "Test 4: Edge Cases FAILED"
|
||
test_results+=("FAIL")
|
||
fi
|
||
((test_count++))
|
||
|
||
# Test 5: Memory Efficiency
|
||
log_info "Test 5/5: Memory Efficiency (10,000+ bars without OOM)..."
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_memory_efficiency --release ${test_args}; then
|
||
log_success "Test 5: Memory Efficiency PASSED"
|
||
test_results+=("PASS")
|
||
((pass_count++))
|
||
else
|
||
log_error "Test 5: Memory Efficiency FAILED"
|
||
test_results+=("FAIL")
|
||
fi
|
||
((test_count++))
|
||
|
||
local end_time=$(date +%s)
|
||
local elapsed=$((end_time - start_time))
|
||
|
||
# Generate summary report
|
||
log_step "Test Summary"
|
||
|
||
echo "Test Results:"
|
||
echo " • Test 1 (Full Pipeline): ${test_results[0]}"
|
||
echo " • Test 2 (Timestamp Alignment): ${test_results[1]}"
|
||
echo " • Test 3 (Performance): ${test_results[2]}"
|
||
echo " • Test 4 (Edge Cases): ${test_results[3]}"
|
||
echo " • Test 5 (Memory Efficiency): ${test_results[4]}"
|
||
echo ""
|
||
echo "Summary:"
|
||
echo " • Total tests: ${test_count}"
|
||
echo " • Passed: ${pass_count}"
|
||
echo " • Failed: $((test_count - pass_count))"
|
||
echo " • Pass rate: $((pass_count * 100 / test_count))%"
|
||
echo " • Total time: ${elapsed}s"
|
||
echo ""
|
||
|
||
if [[ "${pass_count}" -eq "${test_count}" ]]; then
|
||
log_success "ALL TESTS PASSED ✅"
|
||
return 0
|
||
else
|
||
log_error "SOME TESTS FAILED ❌"
|
||
return 1
|
||
fi
|
||
}
|
||
|
||
run_quick_validation() {
|
||
log_step "Running Quick Validation"
|
||
|
||
log_info "Quick validation mode: Running only essential tests..."
|
||
|
||
# Run only Test 1 (Full Pipeline) and Test 3 (Performance)
|
||
local test_args=""
|
||
if [[ "${VERBOSE}" == "true" ]]; then
|
||
test_args="-- --nocapture --test-threads=1"
|
||
fi
|
||
|
||
local pass_count=0
|
||
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_full_replay_pipeline --release ${test_args}; then
|
||
log_success "Full Pipeline test PASSED"
|
||
((pass_count++))
|
||
else
|
||
log_error "Full Pipeline test FAILED"
|
||
fi
|
||
|
||
if cargo test -p ml --test dqn_replay_full_pipeline_test test_replay_performance --release ${test_args}; then
|
||
log_success "Performance test PASSED"
|
||
((pass_count++))
|
||
else
|
||
log_error "Performance test FAILED"
|
||
fi
|
||
|
||
if [[ "${pass_count}" -eq 2 ]]; then
|
||
log_success "Quick validation PASSED ✅"
|
||
return 0
|
||
else
|
||
log_error "Quick validation FAILED ❌"
|
||
return 1
|
||
fi
|
||
}
|
||
|
||
# ============================================================================
|
||
# Cleanup
|
||
# ============================================================================
|
||
|
||
cleanup_temp_files() {
|
||
if [[ "${CLEANUP}" == "true" ]]; then
|
||
log_step "Cleanup"
|
||
log_info "Removing temporary test artifacts..."
|
||
|
||
# Clean up any temporary checkpoints in /tmp
|
||
if ls /tmp/dqn_*.safetensors 1> /dev/null 2>&1; then
|
||
rm -f /tmp/dqn_*.safetensors
|
||
log_success "Removed temporary checkpoints"
|
||
fi
|
||
|
||
# Clean up cargo test artifacts
|
||
cargo clean --release -p ml 2>/dev/null || true
|
||
log_success "Cleaned cargo artifacts"
|
||
fi
|
||
}
|
||
|
||
# ============================================================================
|
||
# Main
|
||
# ============================================================================
|
||
|
||
main() {
|
||
# Parse command-line arguments
|
||
while [[ $# -gt 0 ]]; do
|
||
case $1 in
|
||
--quick)
|
||
RUN_MODE="quick"
|
||
shift
|
||
;;
|
||
--verbose|-v)
|
||
VERBOSE=true
|
||
shift
|
||
;;
|
||
--ci)
|
||
CI_MODE=true
|
||
# Disable colors in CI mode
|
||
GREEN=""
|
||
RED=""
|
||
YELLOW=""
|
||
BLUE=""
|
||
CYAN=""
|
||
BOLD=""
|
||
NC=""
|
||
shift
|
||
;;
|
||
--no-cleanup)
|
||
CLEANUP=false
|
||
shift
|
||
;;
|
||
--help|-h)
|
||
echo "Usage: $0 [OPTIONS]"
|
||
echo ""
|
||
echo "Options:"
|
||
echo " --quick Run quick validation only (2 tests)"
|
||
echo " --verbose, -v Enable verbose logging"
|
||
echo " --ci CI/CD mode (no ANSI colors)"
|
||
echo " --no-cleanup Skip cleanup of temporary files"
|
||
echo " --help, -h Show this help message"
|
||
echo ""
|
||
echo "Examples:"
|
||
echo " $0 # Run all tests"
|
||
echo " $0 --quick # Quick validation"
|
||
echo " $0 --verbose # Verbose output"
|
||
echo " $0 --ci # CI/CD mode"
|
||
exit 0
|
||
;;
|
||
*)
|
||
log_error "Unknown option: $1"
|
||
echo "Use --help for usage information"
|
||
exit 2
|
||
;;
|
||
esac
|
||
done
|
||
|
||
# Print banner
|
||
print_banner
|
||
|
||
# Log configuration
|
||
log_info "Run mode: ${RUN_MODE}"
|
||
log_info "Verbose: ${VERBOSE}"
|
||
log_info "CI mode: ${CI_MODE}"
|
||
echo ""
|
||
|
||
# Run pre-flight checks
|
||
check_dependencies
|
||
|
||
# Run tests based on mode
|
||
local test_result=0
|
||
if [[ "${RUN_MODE}" == "quick" ]]; then
|
||
run_quick_validation || test_result=$?
|
||
else
|
||
run_integration_tests || test_result=$?
|
||
fi
|
||
|
||
# Cleanup
|
||
cleanup_temp_files
|
||
|
||
# Final summary
|
||
if [[ "${test_result}" -eq 0 ]]; then
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo ""
|
||
echo -e "${GREEN}${BOLD}╔══════════════════════════════════════════════════════════════════════╗${NC}"
|
||
echo -e "${GREEN}${BOLD}║ ALL TESTS PASSED ✅ ║${NC}"
|
||
echo -e "${GREEN}${BOLD}╚══════════════════════════════════════════════════════════════════════╝${NC}"
|
||
else
|
||
echo ""
|
||
echo "==================================================================="
|
||
echo " ALL TESTS PASSED"
|
||
echo "==================================================================="
|
||
fi
|
||
exit 0
|
||
else
|
||
if [[ "${CI_MODE}" == "false" ]]; then
|
||
echo ""
|
||
echo -e "${RED}${BOLD}╔══════════════════════════════════════════════════════════════════════╗${NC}"
|
||
echo -e "${RED}${BOLD}║ TESTS FAILED ❌ ║${NC}"
|
||
echo -e "${RED}${BOLD}╚══════════════════════════════════════════════════════════════════════╝${NC}"
|
||
else
|
||
echo ""
|
||
echo "==================================================================="
|
||
echo " TESTS FAILED"
|
||
echo "==================================================================="
|
||
fi
|
||
exit 1
|
||
fi
|
||
}
|
||
|
||
# Run main function
|
||
main "$@"
|