Journal of Causal Inference 6 (2), 2018.
TensorFlow implementation of 'CoRe' (COnditional Variance REgularization),
proposed in "Conditional Variance Penalties and Domain Shift Robustness".
The aim is to build classifiers that are robust against specific interventions.
These domain-shift interventions are defined in a causal graph, extending the
framework of Gong et al (2016). In contrast to Gong et al. we work on a
setting where the domain variable itself is latent but we can observe for
some instances a so-called identifier variables that indicates, for example,
presence of the same person or object across different images. Penalizing the
variance of the predictions across instances that share the same class label and
identifier leads to robustness against strong domain-shift interventions.
Code for 'nonlinear Invariant Causal Prediction' to estimate the
causal parents of a given target variable from data collected in
different experimental or environmental conditions, extending
'Invariant Causal Prediction' from Peters, Buehlmann and Meinshausen (2016)
to nonlinear settings.
Code for a variety of nonlinear conditional independence tests:
Kernel conditional independence test (Zhang et al., UAI 2011),
Residual Prediction test (based on Shah and Buehlmann, arXiv 2015),
Invariant environment prediction,
Invariant target prediction,
Invariant residual distribution test,
Invariant conditional quantile prediction (all from Heinze-Deml et al., arXiv:1706.08576).
Unified interface for the estimation of causal networks, including the methods 'backShift', 'bivariateANM' (bivariate additive noise model),
'bivariateCAM' (bivariate causal additive model), 'CAM' (causal additive model), 'hiddenICP' (invariant causal prediction with hidden variables),
'ICP' (invariant causal prediction), 'GES' (greedy equivalence search), 'GIES' (greedy interventional equivalence search), 'LINGAM',
'PC' (PC Algorithm), 'RFCI' (really fast causal inference) and regression.
LOCOlib implements the LOCO and DUAL-LOCO algorithms for distributed statistical estimation.