v0.2.1

RetentionParameterEstimator v0.2.1

Diff since v0.2.0

Fixed

• Critical performance bug in Loss.jl: Moved unique(substance_list) computation outside the loop in the loss function
  • Dramatically improved performance for single-substance optimizations (method_m1, method_m2, method_m4)
  • Root cause: unique() was being called inside the optimization loop, causing redundant computations during every loss function evaluation
• Test performance: Reduced CI test time from ~71 minutes to ~17 minutes
  • Reduced optimization limits for tests (maxiters=500, maxtime=30.0s instead of defaults)
  • Changed se_col=true to se_col=false for most tests
  • Created separate testset for standard errors with reduced limits
  • Commented out expensive multi-solute test
• Documentation build failure: Fixed "Cannot resolve @ref" errors for stderror function
  • Replaced @ref links with plain backticks to avoid naming conflict with Statistics.stderror
  • Documentation now builds successfully on GitHub Actions

Changed

• Dependencies: Updated GasChromatographySimulator from 0.4/0.5 to 0.6.0
  • Removed Interpolations.jl dependency
  • Now uses GasChromatographySimulator.linear_interpolation and GasChromatographySimulator.deduplicate_knots!
  • Updated reference_holdup_time and estimate_start_parameter_single_ramp to use new interpolation functions
  • Reduced external dependencies, improved consistency with GasChromatographySimulator
• CI and Documentation workflows: Updated GitHub Actions workflows
  • Updated actions/checkout@v2 to @v4 in documentation workflow
  • Updated julia-actions/setup-julia@latest to @v1 in documentation workflow
  • Updated actions/checkout@v3 to @v4 in CI workflow
  • Updated julia-actions/setup-julia@v1 to latest version in CI workflow
  • Updated cache and build actions to latest versions
• Default ODE solver: Changed from OwrenZen5() to Tsit5() in std_opt
  • Better performance and reliability for optimization methods
  • Users can still override by passing custom opt options
• method_m4 convergence check: Added convergence check for substance parameters (Tchar, θchar, ΔCp)
  • Now checks convergence of both d and substance parameters before stopping
  • Uses relative tolerance for substance parameters to handle different scales
  • Prevents premature convergence when only d has converged
  • Improves reliability, especially with Tsit5 ODE solver
• Benchmark scripts: Combined parallel and non-parallel benchmark scripts into single script
  • Added command-line control for parallelization (--parallel, --no-parallel)
  • Added warm-up run to mitigate JIT compilation effects
  • Added git repository state tracking in output files
  • Timestamped output files with parallelization status
  • Simplified output format (summary in .txt, full results in .csv)
• Dependencies: Pinned Optimization.jl to version 5.0.0
  • Versions 5.1.0-5.4.0 caused severe performance regressions (method_m1: <100s → >3600s)
• Documentation: Updated docs/make.jl configuration
  • Specified modules = [RetentionParameterEstimator] for Documenter
  • Added DocMeta.setdocmeta! setup for doctest environment
  • Temporarily disabled doctests to unblock build

Added

• Parallelization support: Added optional parallelization for all optimization methods
  • New parallel=true keyword argument for method_m1, method_m2, method_m3, method_m4
  • Parallelized Kcentric_single and dKcentric_single modes in estimate_parameters
  • Parallelized standard error calculations in stderror and stderror_m3
  • Expected speedup: 4-8x on 4-8 core machines
  • Requires Julia to be started with multiple threads (e.g., julia -t 4)
• Data preparation helper functions: New functions to reduce code duplication
  • prepare_optimization_data(): Consolidates time unit conversion, data extraction, missing value filtering, and vectorization
  • prepare_single_substance_data(): Handles data preparation for single-substance optimization modes
  • time_unit_conversion_factor(): Standardizes time unit conversion
  • Updated prog_list() to handle both single program and vector of programs
• New optimization mode: Added mode="d_only" to estimate_parameters
  • Optimizes only column diameter d while keeping retention parameters fixed
  • Integrated into method_m4 alternating optimization strategy
• New benchmark script: scripts/benchmark_m1.jl for dedicated method_m1 benchmarking
• Convergence tracking: Added convergence checks for substance parameters in method_m4
  • Tracks previous Tchar, θchar, ΔCp values across iterations
  • Requires convergence of all parameters before stopping optimization

Breaking

None. All changes are backward compatible.

Merged pull requests:

• Update gas chromatograhy simulator to 0 5 7 (#41) (@JanLeppert)
• Update ci and documentation workflow (#42) (@JanLeppert)
• New optimization method (#43) (@JanLeppert)
• added helper functions (#44) (@JanLeppert)
• Add parallelization (#45) (@JanLeppert)
• Fix optimization speed (#46) (@JanLeppert)
• Resolve documentation build failure on GitHub Actions (#47) (@JanLeppert)
• Update v0.2.1 (#48) (@JanLeppert)