Information-Theoretic Causal Inference on Discrete Data
MIT License
Caddie is a collection of bivariate discrete causal inference methods based on information-theoretic Additive Noise Models (ANM) and MDL-based instantiation of Algorithmic Independence of Conditionals (AIC).
The recommended way to install the caddie
module is to simply use pip
:
$ pip install caddie
Caddie officially supports Python >= 3.6.
>>> X = [1] * 1000
>>> Y = [-1] * 1000
>>> from caddie import cisc
>>> cisc.cisc(X, Y) # CISC
(0.0, 0.0)
>>> from caddie import anm, measures
>>> anm.fit_both_dir(X, Y, measures.StochasticComplexity) # CRISP
(0.0, 0.0)
>>> anm.fit_both_dir(X, Y, measures.ChiSquaredTest) # DR
(1.0, 1.0)
>>> anm.fit_both_dir(X, Y, measures.ShannonEntropy) # ACID
(0.0, 0.0)
>>> from caddie import simulations
>>> simulations.simulate_decision_rate_against_data_type('/results/dir/') # for decision rate vs data type plots
...
>>> simulations.simulate_accuracy_against_sample_size('/results/dir/') # for accuracy/decidability vs sample size plots
...
Todo: Add the citation to thesis.