Luminator

Simplified XAI framework

Luminator delivers the computation of all kinds of machine learning explanations with a simple, yet powerful API.

Example Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer
from luminator.algorithms.deeplift import DeepLiftParameters
from luminator.luminators.nlp.classification import SequenceClassificationLuminator

# load transformer model
model_name = "distilbert-base-uncased-finetuned-sst-2-english"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# initialize luminator
luminator = SequenceClassificationLuminator(model=model, tokenizer=tokenizer)

# define parameters for desired algorithm
params = DeepLiftParameters(
    examples=["Luminator is a great tool for explainability"],
    targets=[1, 1, 1],
    batch_size=2
)

# generate explanations
explanations = luminator.deeplift(params=params)

Installation

The library is not published yet. Therefore it must be installed from source.

1. Clone the repository

   git clone https://github.com/fmaoro/luminator.git
   

2. Install with pip

   pip install .
   

Documentation

See documentation for details