moonsplit

    A MoonBit dataset grouping, splitting, and leakage audit library for reproducible machine learning workflows.

    dataset
    split
    leakage
    ml
    research
    Download zip
    Version
    0.1.1
    License
    Apache-2.0
    Last updated
    5 hours ago
    Downloads
    2

    Dependencies

    #MoonSplit

    MoonSplit is a MoonBit CLI and library for grouped dataset splitting and leakage auditing. It keeps records linked by declared metadata keys in the same partition, produces deterministic split plans, and independently audits the resulting assignment.

    #Features

    • Multi-key isolation with transitive grouping.
    • Deterministic holdout and k-fold splitting.
    • Independent audit from raw records and assignment manifest.
    • Stable JSON, JSONL, and Markdown exports.
    • wasm-gc and JS targets.

    #Installation

    moon add AlexenderSokolov/moonsplit

    To run from source:

    git clone https://github.com/AlexenderSokolov/moonsplit.git cd moonsplit

    The examples/ directory contains ready-made JSON, JSONL, and split specs.

    #Quick Start

    ./run_check.ps1 ./run_demo.ps1

    ./run_check.sh ./run_demo.sh ./run_bench.sh

    #Commands

    moon run cmd/moonsplit -- plan --data examples/voice_records.jsonl --spec examples/holdout_spec.json --out examples/demo_run moon run cmd/moonsplit -- audit --data examples/voice_records.jsonl --spec examples/holdout_spec.json --assignments examples/demo_run/assignments.jsonl moon run cmd/moonsplit -- demo --out examples/demo_run moon run cmd/moonsplit -- --version moon run cmd/moonsplit -- smoke --data examples/voice_records.jsonl

    The output directory must not already exist. The parent directory must exist. Exit codes are:

    • 0: success
    • 2: usage, input, or read error
    • 3: audit failure
    • 4: output write error

    smoke reads one input file and prints its character count for quick file-system checks.

    The output files are documented in docs/output.md.

    #Scope

    MoonSplit does not do time-series splitting, raw media similarity detection, fingerprinting, training, scheduling, databases, web UIs, or CSV frameworks. It does not claim that data are completely leak-free; it only audits the declared metadata isolation rules.

    Powered by MoonBit

    Site sourceReport issuePackagesBuild queueSkillsStatistics

    © 2026 mooncakes.io