Regexulator: regex from examples using LLMs

Uses Ollama to generate regular expression patterns from examples.

Usage

from regexulator import Regexulator

examples = [{
  "string": "It will start at 10:30 and end at 11:21.",
  "match": [
    {"start": 17, "end": 22},
    {"start": 34, "end": 39},
  ]
}]

reg = Regexulator(model="llama3.1:70b", time_limit=10)
pattern = reg.fit(examples)

Configuration

You can use the RegexulatorConfig to make a reusable config with type hints. Keyword arguments supplied directly to Regexulator overwrite individual config attributes.

from regexulator import Regexulator, RegexulatorConfig

cfg = RegexulatorConfig(
    model="llama3.1:70b",
    time_limit=60,
    initial_splits=4,
    max_depth=5,
    branching_factor=1
)
reg1 = Regexulator(cfg, time_limit=10)
reg2 = Regexulator(cfg, time_limit=200, max_depth=8)

.dot visualisation

You can create a graphviz .dot representation of the exploration tree with the Regexulator.to_dot() method.

In a Jupyter notebook, you can just call Regexulator.show_dot() (this needs the graphviz module)