Md. Asif Uddin

    Book VI

    Causal inference

    DAGs, interventions, counterfactuals, perturbation

    Move from prediction to intervention: what a causal claim is, when data can support one, and when it cannot.

    9 chapters0 propositions written

    Read first

    neural networks

    Mathematics assumed — follow these when a step stops making sense.

    Conditional independence 0.PR.05 · Bayes' rule 0.PR.04 · Estimators, bias and variance 0.ST.01 · Hypothesis tests 0.ST.03

    1. Chapter 1VI.1Correlation and CausationThe same joint distribution can be produced by different causal structures, so data alone cannot decide between them.Correlation · Causation · Confounding · Prediction versus interventionM2not yet written

      0/3 problems0/3 variants0/6 exercisesowes 9 more

    2. Chapter 2VI.2Causal GraphsA causal graph encodes conditional independences, and d-separation reads them off the graph alone.DAGs · Nodes · Edges · Paths · Ancestors · Descendants · Colliders · ConfoundersM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    3. Chapter 3VI.3InterventionsAn intervention deletes the mechanism that generated a variable, and conditioning does not.Do-operator · Intervention · Observational distribution · Interventional distributionM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    4. Chapter 4VI.4IdentificationIdentification asks whether an interventional quantity can be written in terms of observed ones, and the graph answers it.Backdoor criterion · Adjustment · Front-door criterion · Conditional independenceM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    5. Chapter 5VI.5CounterfactualsA counterfactual asks about an individual outcome that was never observed, and no experiment can supply it.Potential outcomes · Counterfactual worlds · Individual treatment effects · Causal effectM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    6. Chapter 6VI.6ExperimentsRandomisation buys identification, and the price is stated in advance as a sample size.Randomisation · Treatment and control · A/B testing · Experimental design · Power · BiasM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    7. Chapter 7VI.7Observational StudiesWithout randomisation every assumption must be argued, and positivity is the one that fails quietly.Propensity scores · Matching · Inverse probability weighting · Regression adjustment · Sensitivity analysisM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    8. Chapter 8VI.8Causal Machine LearningMachine learning estimates the nuisance functions, and the estimator's structure is what makes the effect robust to getting them wrong.Heterogeneous treatment effects · Causal forests · Representation learning · Neural causal models · Causal discoveryM3not yet written

      0/5 problems0/4 variants0/10 exercisesowes 15 more

    9. Chapter 9VI.9Causal Reasoning in AIA spurious feature is a confounder with a different name, and the vocabulary of this Book applies to models as it does to treatments.Machine learning · Vision · Language · Scientific discovery · HealthcareM2not yet written

      0/3 problems0/3 variants0/6 exercisesowes 9 more

    Problem setWhere the chapters have to be used togetherNot yet written. A book with load-bearing chapters owes at least eight cross-chapter problems and two that reach back into an earlier book.

    Practical connection

    DAG analysis, then ATE estimation, then a synthetic intervention.

    The adjusted effect your code returns must equal the one you computed from the 2x2x2 table.

    Verified againstVI.4.B02 · VI.7.B02