Thanks for sharing. What are the areas where this approach would provide the most benefit Vs SCMs and Bayesian Networks? What are the trade-offs? Linking to t

Thanks for sharing. What are the areas where this approach would provide the most benefit Vs SCMs and Bayesian Networks? What are the trade-offs? Linking to their tutorial notebook: https://colab.research.google.com/github/deepmind/deepmind_research/blob/master/causal_reasoning/Causal_Reasoning_in_Probability_Trees.ipynb https://colab.research.google.com/github/deepmind/deepmind_r... Questions: 1. Has anyone used a package like "bnlearn" in R to guide the structure learning of one of these starting from data? 2. Would be curious about their relationship to CART models, Random Forest models and Causal Decision Trees (http://nugget.unisa.edu.au/jiuyong/CDT-Revision.pdf http://nugget.unisa.edu.au/jiuyong/CDT-Revision.pdf) 3. This feels like an important limitation Vs SCMs: "A system of transformations akin to the do-calculus is entirely absent and left for future work. Such a system is necessary for addressing questions about identification, i.e. whether a causal effect can be estimated from observation alone"