generate

Description

Generate synthetic BED datasets for benchmarking. Creates reproducible test data with various distribution patterns.

Command

grit generate --output ./benchmark_data --sizes 1M,5M,10M --seed 42

Output

Creates directory structure with A.bed and B.bed files:

benchmark_data/
├── balanced/
│   ├── 1M/
│   │   ├── A.bed
│   │   └── B.bed
│   ├── 5M/
│   └── 10M/
├── clustered/
│   └── ...
├── identical/
│   └── ...
├── skewed-a-gt-b/
│   └── ...
└── skewed-b-gt-a/
    └── ...

Options

Flag	Description
-o, --output	Output directory (default: ./grit_bench_data)
--sizes	Sizes to generate (default: 1M,5M,10M,25M,50M)
--seed	Random seed for reproducibility (default: 42)
--mode	Generation mode (default: all)
--sorted	Sorting behavior: yes, no, auto (default: auto)
--no-sort	Alias for --sorted no
--a	Custom A file size
--b	Custom B file size
--hotspot-frac	Genome fraction for hotspots (default: 0.05)
--hotspot-weight	Interval fraction in hotspots (default: 0.80)
--len-min	Minimum interval length (default: 50)
--len-max	Maximum interval length (default: 1000)
--force	Overwrite existing files

Generation Modes

Mode	Description
balanced	Equal-sized A and B files with uniform distribution
skewed-a-gt-b	A file 10x larger than B
skewed-b-gt-a	B file 10x larger than A
identical	A and B are identical
clustered	Intervals concentrated in hotspot regions
all	Generate all modes

Size Specifications

Format	Example	Value
Number	1000	1,000 intervals
K suffix	10K	10,000 intervals
M suffix	5M	5,000,000 intervals

Examples

Quick Benchmark Data

grit generate --sizes 100K --mode balanced --force

Large-Scale Testing

grit generate --sizes 10M,50M,100M --mode all --seed 123

Custom A/B Sizes

grit generate --a 1M --b 10M --mode balanced

Clustered Data

grit generate --mode clustered --hotspot-frac 0.1 --hotspot-weight 0.9

Unsorted Output

grit generate --sizes 1M --no-sort

Notes

• All generated files use human genome chromosome sizes (hg38)
• Seed ensures reproducibility across runs
• Auto sort mode sorts when size <= 10M
• Generated data suitable for benchmarking intersect, coverage, etc.