benchmark

Internal Benchmarks

This directory contains pure Julia benchmarks for development and regression testing of ChenSignatures.jl.

It allows maintainers to quickly check the performance of the library without setting up Python environments or external dependencies.

📊 Cross-Language Benchmarks:
For comprehensive comparisons against iisignature (C++) and pysiglib (PyTorch), please see the dedicated benchmark repository:
👉 aleCombi/sig-benchmarks

---

📂 Files

• benchmark.jl: The main script. Runs signatures and log-signatures on generated paths and reports timing/allocations.
• benchmark_config.yaml: Configuration file to control problem sizes ($N, d, m$) and execution parameters.

---

🚀 Usage

To run the benchmarks, you only need a working Julia installation.

1. Instantiate dependencies (only needs to be done once):

   julia --project=benchmark -e 'using Pkg; Pkg.develop(path="."); Pkg.instantiate()'

2. Run the benchmarks:

   julia --project=benchmark benchmark/benchmark.jl

Output

The script will:

1. Print a summary to the console.
2. Create a timestamped folder in runs/ (e.g., runs/benchmark_julia_2025.../) containing:
   • results.csv: Detailed timing and allocation data for every configuration.
   • SUMMARY.txt: A quick overview of the run.

---

⚙️ Configuration

Edit benchmark_config.yaml to customize the sweep:

# Grid parameters
Ns: [1000, 10000]      # Path lengths
Ds: [3, 10]            # Dimensions
Ms: [4, 6]             # Signature depths

# Options
path_kind: "sin"       # "linear" or "sin"
repeats: 5             # Samples per test

Tip: Keep the grid small for quick regression checks during development.