Handy Lexicon
List of pure Bayes lexicon is in https://statmodeling.stat.columbia.edu/2009/05/24/handy_statistic/
- sin vs exponential grwoth?
- multimodality might be not relevant to your problem
- to prevent gardens of forking path in modeling hetereogenity (relevant to Best cluster may not come from clustering)
- wrongly set upper bound of parameter hampering the estimation by cutting off the head
- Using table function as loop's abstraction: functional mapping (SW vs base point parameter
- Tom is also using "parameterize table functions for sensitivity testing" for his deer chronic disease aging chain, but Tom is experiencing parameter's upper bound acting as a ceiling
- relevant to Loop knockout, causal identification as described in
- somewhere in between
- Tom showed me the picture of waterfall
- only nature can produce full fractal
- screen function
- 0_PA, 1_PAD, 2_PD, 3_Data4DM, 4_DM4Data
- labeled then unlabeled
- descriptive
- visualization
- clustering
- latent variable identification & generative approaches
- Bayesian (theory-based): HMM, Particle filtering, PMCMC
- Connectionist (less theory-based): Autoencoders, VAEs, GANs
- Dimensionality reduction (PCA/ICA, t-sne, SVC)
- Causality detection (CCM)
- Density estimation
- regression
- classification key: defining loss function and regularization
- correlation doesn't imply causation
- seeks to rigorusly predict outcomes in accordance w/posited causal structure
- advantage
- capacity to reason about counterfactuals
- strong generalizability across contexts
- enhanced explainability
- heavy reliance upon postulated causal structure
- can cross-check causal expectations using empirical data via conditional independence, reverse dependence
- in temporal settings, causal hints can be suggested by empirical data (CCM)
Nathaniel explains the above as: Description (unsupervised), Prediction (supervised/semi-supervised), Causal prediction (both supervised & unsupervised).