How do science on Dan MacKinlay
https://danmackinlay.name/tags/how_do_science.html
Recent content in How do science on Dan MacKinlayHugo -- gohugo.ioen-usTue, 13 Apr 2021 16:24:05 +0800Machine learning for partial differential equations
https://danmackinlay.name/notebook/ml_pde.html
Tue, 13 Apr 2021 16:24:05 +0800https://danmackinlay.name/notebook/ml_pde.htmlLearning a PDE Deterministic PINN Stochastic PINN Weak formulation Learning a PDE forward operator Fourier neural operator DeepONet Advection-diffusion PDEs in particular Boundary conditions Inverse problems Differentiable solvers DeepXDE ADCME TenFEM JuliaFEM Trixi FEniCS taichi References Using statistical or machine learning approaches to solve PDEs, and maybe even to perform inference through them. There are various approaches here, which I will document on an ad hoc basis as I need them.Differentiable model selection
https://danmackinlay.name/notebook/model_selection_diff.html
Tue, 13 Apr 2021 15:56:46 +0800https://danmackinlay.name/notebook/model_selection_diff.htmlReferences Maclaurin, Duvenaud, and Adams (2015):
Hyperparameter optimization by gradient descent
Each meta-iteration runs an entire training run of stochastic gradient de- scent to optimize elementary parameters (weights 1 and 2). Gradients of the validation loss with respect to hyperparameters are then computed by propagating gradients back through the elementary training iterations. Hyperparameters (in this case, learning rate and momentum schedules) are then updated in the direction of this hypergradient.Machine learning for physical sciences
https://danmackinlay.name/notebook/ml_physics.html
Fri, 09 Apr 2021 11:48:30 +0800https://danmackinlay.name/notebook/ml_physics.htmlML for PDEs Causality, identifiability, and observational data Likelihood free inference Emulation approaches The other direction: What does physics say about learning? But statistics is ML Applications References Consider a spherical flame
In physics, typically, we are concerned with identifying True Parameters for Universal Laws, applicable without prejudice across all the cosmos. We are hunting something like the Platonic ideals that our experiments are poor shadows of.Scientific writing
https://danmackinlay.name/notebook/scientific_writing.html
Tue, 30 Mar 2021 12:18:37 +1100https://danmackinlay.name/notebook/scientific_writing.htmlProcess Review and rebuttals References References Is there a recommended style guide for theses? Not formatting, but linguistic style? I’m presuming there must be one, or they would not generally be written so horribly. I would like to know where this gaggingly awful style is elucidated so that I may ape it and thereby adhere to the expectations of this noble institution that is the thesis, and also maybe work out what small freedoms I may have to write comprehensibly, interesting etc.Science; Sociology and institution design for
https://danmackinlay.name/notebook/peer_review.html
Tue, 30 Mar 2021 11:19:44 +1100https://danmackinlay.name/notebook/peer_review.htmlOpen review processes, practical Mathematical models of the reviewing process Economics of publishing Mechanism desoign for peer review process How well does academia gatekeep? Style guide for reviews and rebuttals References Upon the thing that I presume academic publishing is supposed to do: further science. Reputation system and other mechanisms for trust in science, a.k.a. collective knowledge for reality itself.
I would like to consider the system of peer review, networking, conferencing, publishing and acclaim and see how closely it approximates an ideal system for uncovering truth, and further, imagine how we could make a better system.Filter bubbles, fact checking and kompromat
https://danmackinlay.name/notebook/filter_bubbles.html
Sun, 21 Mar 2021 15:51:21 +1100https://danmackinlay.name/notebook/filter_bubbles.htmlReferences News media and public shared reality. Fake news, incomplete news, alternative facts, strategic inference, kompromat, agnotology, facebooking to a molecular level. Basic media literacy and whether it helps. As seen in elections, and provocateur twitter bots.
Theirtube
Theirtube is a Youtube filter bubble simulator that provides a look into how videos are recommended on other people’s YouTube. Users can experience how the YouTube home page would look for six different personas.BibLaTeX
https://danmackinlay.name/notebook/biblatex.html
Thu, 18 Mar 2021 10:20:42 +1100https://danmackinlay.name/notebook/biblatex.htmlBibTeX BibLaTeX How citation management works in LaTeX.
BibTeX Oh, BibTeX. The classic LaTeX-compatible citation system.
None of that faffing about making web-friendly citations is useful if you are working with academics, who don’t regard words on the internet as a real thing. Your words must be behind a paywall where no-one can read them to count as significant. Moreover, they must have been rendered harder to analyse by running them through LaTeX to obfuscate them into a PDF, which probably also entails using BibTeX to do the citation stuff.Probabilistic programming
https://danmackinlay.name/notebook/probabilistic_programming.html
Wed, 10 Mar 2021 13:59:55 +1100https://danmackinlay.name/notebook/probabilistic_programming.htmlFunsors MCMC considerations Variation inference considerations Toolkits Stan Pyro Numpyro Edward/Edward2 TensorFlow Probability pyprob Turing.jl PyMC3 Mamba.jl Gen Greta Soss.jl Miscellaneous julia options Inferpy Zhusuan Church/Anglican WebPPL BAT References Programming with random control paths, where the goal is to estimate the probability of ended up on a certain state, conditional upon some input.
More specifically, what I usually use probabilistic programming for is Bayesian inference, and I think this is common enough that it is generally assumed.Blogdown
https://danmackinlay.name/notebook/blogdown.html
Wed, 10 Mar 2021 08:53:01 +1100https://danmackinlay.name/notebook/blogdown.htmlBlogdown classic hugodown Pain points distill + radix Blogdown is the academic blogging system that I use. It is a system to construct a website using pages made from rmarkdown and similar literate programming tools to write a blog with embedded diagrams and code and such. In its default state it just works, and is highly recommended. However, after a while there are a few frictions that one can avoid by being fancy (e.knitr/RMarkdown
https://danmackinlay.name/notebook/knitr_rmarkdown.html
Wed, 03 Mar 2021 12:35:47 +1100https://danmackinlay.name/notebook/knitr_rmarkdown.htmlPro tips RMarkdown Tips Customizing pandoc Tables Editor support Citations knitr is the R-based entrant in the scientific workbook race, combining the code that creates your data analysis with the text, keeping them in sync in perpetuity.
It is frequently used in the form of RMarkdown which supports the markdown input format instead of e.g. LaTeX. I most often use it in the form of blogdown, which is the engine that drives this blog.Convolutional Gaussian processes
https://danmackinlay.name/notebook/gp_convolution.html
Mon, 01 Mar 2021 17:08:51 +1100https://danmackinlay.name/notebook/gp_convolution.htmlConvolutions with respect to a non-stationary driving noise Varying convolutions with respect to a stationary white noise References Gaussian processes by convolution of noise with smoothing kernels, which is a kind of dual to defining them through covariances.
This is especially interesting because it can be made computationally convenient (we can enforce locality) and non-stationarity.
Convolutions with respect to a non-stationary driving noise H. K.Convolutional stochastic processes
https://danmackinlay.name/notebook/stochastic_convolution.html
Mon, 01 Mar 2021 16:13:24 +1100https://danmackinlay.name/notebook/stochastic_convolution.htmlReferences Stochastic processes generated by convolution of white noise with smoothing kernels, which is not unlike kernel density estimation where the “data” is random.
For now, I am mostly interested in certain special cases Gaussian process convolutionss and subordinator convolutions.
patrick-kidger/Deep-Signature-Transforms: Code for "Deep Signature Transforms" patrick-kidger/signatory: Differentiable computations of the signature and logsignature transforms, on both CPU and GPU. References Bolin, David.Hyperparameter optimization in ML
https://danmackinlay.name/notebook/hyperparam_opt.html
Mon, 01 Mar 2021 09:10:44 +1100https://danmackinlay.name/notebook/hyperparam_opt.htmlBayesian/surrogate optimisation Differentiable hyperparameter optimisation Random search Adaptive random search Implementations Determined Ray Optuna hyperopt auto-sklearn skopt spearmint SMAC AutoML References Split off from autoML.
The art of choosing the best hyperparameters for your ML model’s algorithms, of which there may be many.
Should you bother getting fancy about this? Ben Recht argues no, that random search is competitive with highly tuned Bayesian methods in hyperparameter tuning.Causal inference in the continuous limit
https://danmackinlay.name/notebook/causality_continuous.html
Wed, 17 Feb 2021 20:41:08 +1100https://danmackinlay.name/notebook/causality_continuous.htmlReferences Causality on continuous index spaces, and, which turns out to be related, equilibrium/feedback dynamics. Placeholder.
Bongers and Mooij (2018):
Uncertainty and random fluctuations are a very common feature of real dynamical systems. For example, most physical, financial, biochemical and engineering systems are subjected to time-varying external or internal random disturbances. These complex disturbances and their associated responses are most naturally described in terms of stochastic processes.Bayesians vs frequentists
https://danmackinlay.name/notebook/bayesians_vs_frequentists.html
Thu, 14 Jan 2021 15:35:59 +1100https://danmackinlay.name/notebook/bayesians_vs_frequentists.htmlAvoiding the whole accursed issue Frequentist vs Bayesian acrimony Strong Bayesianism References Disagreements in posterior updates
Sundry schools thought in how to stitch mathematics to the world, brief notes and questions thereto. Justin Domke wrote a Dummy’s guide to risk and decision theory which explains the different assumptions underlying each methodology from the risk and decision theory angle.
A lot of the obvious debates here are, IMO, uninteresting.Multi-output Gaussian process regression
https://danmackinlay.name/notebook/gp_regression_functional.html
Mon, 07 Dec 2020 20:43:06 +1100https://danmackinlay.name/notebook/gp_regression_functional.htmlReferences In which I discov Learning operators via GPs.
References Brault, Romain, Florence d’Alché-Buc, and Markus Heinonen. 2016. “Random Fourier Features for Operator-Valued Kernels.” In Proceedings of The 8th Asian Conference on Machine Learning, 110–25. http://arxiv.org/abs/1605.02536. Brault, Romain, Néhémy Lim, and Florence d’Alché-Buc. n.d. “Scaling up Vector Autoregressive Models With Operator-Valued Random Fourier Features.” Accessed August 31, 2016. https://aaltd16.irisa.fr/files/2016/08/AALTD16_paper_11.pdf. Brouard, Céline, Marie Szafranski, and Florence D’Alché-Buc.Crowd-sourced science
https://danmackinlay.name/notebook/citizen_science.html
Tue, 24 Nov 2020 11:17:31 +1100https://danmackinlay.name/notebook/citizen_science.htmlGame theory of crowdsourced competitions Tools References Mapping the world from smartphones. Buzzwords: Citizen science
LIMN magazine on crowds and clouds Game theory of crowdsourced competitions Crowd sourcing competitions encourage malicious behaviour Game theory of crowdsource competition Tools Software frameworks for data collection: (Thanks Dan Pagendam for the tip.)
ODK-X ODK - Collect data anywhere References Charness, Gary, and Matthias Sutter.External validity
https://danmackinlay.name/notebook/external_validity.html
Mon, 09 Nov 2020 15:58:56 +1100https://danmackinlay.name/notebook/external_validity.htmlStandard graphical models Tools Salad Meta References TBD.
This Maori gentleman from the 1800s demonstrates an artful transfer learning from the western fashion domain
One could read Sebastian Ruder’s NN-style introduction to “transfer learning”. NN people like to think about this in particular way which I like because of the diversity of out-of-the-box ideas it invites and which I dislike because it is sloppy.Experiment tracking for computation science
https://danmackinlay.name/notebook/experiment_tracking.html
Fri, 06 Nov 2020 21:03:14 +1100https://danmackinlay.name/notebook/experiment_tracking.htmlSacred ml-metadata Dr Watson Lancet CaosDB Forge Estimator Artemis Pachyderm Sumatra Pathos Ruffus Make. Closely-related problem, in this modern data-driven age: versioning and managing data, build tools.
Sacred Sacred is a tool to configure, organize, log and reproduce computational experiments. It is designed to introduce only minimal overhead, while encouraging modularity and configurability of experiments.
The ability to conveniently make experiments configurable is at the heart of Sacred.Causal inference on DAGs
https://danmackinlay.name/notebook/causal_inference.html
Wed, 04 Nov 2020 12:36:13 +1100https://danmackinlay.name/notebook/causal_inference.htmlLearning materials do-calculus Counterfactuals Continuously indexed fields External validity Propensity scores Causal Graph inference from data Causal time series DAGS Drawing graphical models Tools References Inferring the optimal intervention requires accounting for which arrows are independent of which
Inferring cause and effect from nature. Graphical models and related techniques for doing it. Avoiding the danger of folk statistics. Observational studies, confounding, adjustment criteria, d-separation, identifiability, interventions, moral equivalence…Digital scientific workbooks
https://danmackinlay.name/notebook/scientific_workbooks.html
Tue, 03 Nov 2020 16:07:38 +1100https://danmackinlay.name/notebook/scientific_workbooks.htmlPhilosophy Sharing Glamorous Toolkit Deepnote Jupyter Pweave Weave.jl Tangle Stencila knitr/RMarkdown MATLAB Editor/IDE support Atom VS Code References The exploratory-algorithm-person’s IDE-equivalent. Literate coding-meets-science. a.k.a. dynamic report generation, a.k.a. literate programming.
Let’s say I want to demonstrate my algorithm to my thesis advisor while he’s off at conference. I need an easily shareable demonstration. that’s why we have the internet, right? I should be able to interleave text and mathematics and also code demonstrating the thingy, maybe even some graphs of the output.Publication bias
https://danmackinlay.name/notebook/publication_bias.html
Sun, 01 Nov 2020 15:16:14 +1100https://danmackinlay.name/notebook/publication_bias.htmlFixing P-hacking References We’re out here everyday, doing the dirty work finding noise and then polishing it into the hypotheses everyone loves. It’s not easy. —John Schmidt, The noise miners
The noise mining process
Multiple testing across a whole scientific field, with a side helping of uneven data release.
On one hand we hope that journals will help us find things that are relevant.Stan
https://danmackinlay.name/notebook/stan.html
Mon, 19 Oct 2020 10:06:01 +1100https://danmackinlay.name/notebook/stan.htmlReferences Stan magically reducing posterior distribution surfaces to a smooth traversible manifold
Stan is the inference toolbox for broad classes of Bayesian inference, daaaaahling. Usually seen in concert with brms, which makes it easier to use in various standard regression models, but “bareback” is also pretty simple.
Baked-in documentation is extensive, but there is more. Andrew Gelman notes:
The basic execution structure of Stan is in the JSS paper (by Bob Carpenter, Andrew Matt Hoffman, Daniel Lee, Ben Goodrich, Michael Betancourt, Marcus Brubaker, Jiqiang Guo, Peter Li, and Allen Riddell) and in the reference manual.Automatic design of experiments
https://danmackinlay.name/notebook/design_of_experiments.html
Tue, 13 Oct 2020 17:50:06 +1100https://danmackinlay.name/notebook/design_of_experiments.htmlProblem statement Aquisition functions Connection to RL Implementation skopt Dragonfly PySOT GPyOpt Sigopt BoTorch/Ax spearmint SMAC References Closely related is AutoML, in that surrogate optimisation is a popular tool for such.
Problem statement According to Gilles Louppe and Manoj Kumar:
We are interested in solving
\[x^* = \arg \min_x f(x)\]
under the constraints that
\(f\) is a black box for which no closed form is known (nor its gradients); \(f\) is expensive to evaluate; evaluations of \(y=f(x)\) may be noisy.Reproducible research
https://danmackinlay.name/notebook/reproducible_research.html
Tue, 13 Oct 2020 11:24:08 +1100https://danmackinlay.name/notebook/reproducible_research.htmlActually existing reproducible research tools Peer review Open notebooks Examples of online notebooks Containerized workflow Build tools Sundry data sharing ideas Collaboration Communities and organisations References Philosophies of how to share your methods for the purpose of Doing Science. More technical details are under build/pipelines tools, experiment tracking and scientific workbooks. There should probably be some continuous integration system mentioned here, but maybe later; we are taking baby steps.AutoML
https://danmackinlay.name/notebook/automl.html
Fri, 02 Oct 2020 06:29:47 +1000https://danmackinlay.name/notebook/automl.htmlReinforcement learning approaches Differentiable architecture search Implementations auto-sklearn References The sub-field of optimisation that specifically aims to automate model selection in machine learning. (and also occasionally ensemble construction)
There are two major approaches here that I am aware of, both of which are related in a kind of abstract way, but which are in practice different
Finding the right architecture for your nueral net, a.Research data sharing
https://danmackinlay.name/notebook/data_sharing.html
Wed, 23 Sep 2020 12:51:28 +1000https://danmackinlay.name/notebook/data_sharing.htmlWeb data repositories dolthub DVC Dat Orbitdb Qu Misc Tips and tricks for collaborative data sharing, e.g. for reproducible research.
Related: the problems of organising the data efficiently for the task in hand. For that, see database, and the task of versioning the data. Also related: the problem of finding some good data for your research project, ideally without having to do the research yourself. For some classic datasets that use these data sharing methods (and others) see data sets.Tests, statistical
https://danmackinlay.name/notebook/statistical_tests.html
Mon, 21 Sep 2020 18:57:38 +1000https://danmackinlay.name/notebook/statistical_tests.htmlGoodness-of-fit tests Design of experiments References The mathematics of the last century worth of experiment design. This is about the classical framing, where you think about designing and running experiments and deciding if can reasonably be construed to be true or not, then go home. There are many elaborations of this approach in the modern world. For example, we examine large numbers of hypotheses at once under multiple testing.Metis and .*-rationality
https://danmackinlay.name/notebook/metis_truth.html
Mon, 21 Sep 2020 13:23:36 +1000https://danmackinlay.name/notebook/metis_truth.htmlThe function of belief in individuals Belief and groups The rationality of the Great Society Policy and Statistical learning References Spontaneous order, local knowledge, strategic belief, and other castings of the relationship of beliefs and knowledge in the social order. I have no original thoughts on this, but I like to keep links on this theme where I can see them so that they don’t bite me.Causal inference in highly parameterized ML
https://danmackinlay.name/notebook/causality_ml.html
Fri, 18 Sep 2020 09:34:46 +1000https://danmackinlay.name/notebook/causality_ml.htmlReferences TBD.
Léon Bottou, From Causal Graphs to Causal Invariance
For many problems, it’s difficult to even attempt drawing a causal graph. While structural causal models provide a complete framework for causal inference, it is often hard to encode known physical laws (such as Newton’s gravitation, or the ideal gas law) as causal graphs. In familiar machine learning territory, how does one model the causal relationships between individual pixels and a target prediction?Academic blogging workflow
https://danmackinlay.name/notebook/academic_blogging_workflow.html
Fri, 04 Sep 2020 10:15:31 +1000https://danmackinlay.name/notebook/academic_blogging_workflow.htmlBlogdown hugodown distill + radix Wowchemy Jekyll Pelican Hakyll pandoc-scholar Pollen Madoko Jupyter Hexo b-ber Franklin This blog, and virtually all my notes, are in plain text files on my computer, published online as plain html files. It’s an informal open notebook.
I had to jump through some hoops to make this work, because I need mathematical markup support and basic citation management. Vanilla, non-academic plain text blogging is simpler.Causal graphical model reading group 2020
https://danmackinlay.name/post/reading_group_2020_causal_dags.html
Thu, 03 Sep 2020 11:10:51 +1000https://danmackinlay.name/post/reading_group_2020_causal_dags.htmlMotivational examples Generally Machinery Structural Equation Models Directed Acyclic Graphs (DAGs) Causal interpretation do-calculus Case study: Causal GPs Recommended reading Quick intros Textbooks Questions References See also a previous version, and the notebook on causal inference this will hopefully inform one day.
We follow Pearl’s summary (Pearl 2009a), sections 1-3.
In particular, the paper surveys the development of mathematical tools for inferring (from a combination of data and assumptions) answers to three types of causal queries: (1) queries about the effects of potential interventions, (also called “causal effects” or “policy evaluation”) (2) queries about probabilities of counterfactuals, (including assessment of “regret,” “attribution” or “causes of effects”) and (3) queries about direct and indirect effects (also known as “mediation”)Causal Bayesian networks
https://danmackinlay.name/notebook/causal_bayesian_networks.html
Tue, 01 Sep 2020 08:29:10 +1000https://danmackinlay.name/notebook/causal_bayesian_networks.htmlReferences Some kind of alternative graphical formalism for causal independence graphs 🤷?
discrete probability trees, sometimes also called staged tree models. A probability tree is one of the simplest models for representing the causal generative process of a random experiment or stochastic process The semantics are self-explanatory: each node in the tree corresponds to a potential state of the process, and the arrows indicate both the probabilistic transitions and the causal dependencies between them.Bushfire models
https://danmackinlay.name/notebook/bushfires.html
Wed, 19 Aug 2020 11:34:20 +1000https://danmackinlay.name/notebook/bushfires.htmlReferences american forest fire 1904
Press: Silicon valley wildfire spotting.
References Hilton, J. E., A. L. Sullivan, W. Swedosh, J. Sharples, and C. Thomas. 2018. “Incorporating Convective Feedback in Wildfire Simulations Using Pyrogenic Potential.” Environmental Modelling & Software 107 (September): 12–24. https://doi.org/10.1016/j.envsoft.2018.05.009. Mandel, Jan, Jonathan D. Beezley, Janice L. Coen, and Minjeong Kim. 2009. “Data Assimilation for Wildland Fires.” IEEE Control Systems Magazine 29 (3): 47–65.Citation management
https://danmackinlay.name/notebook/citation_management.html
Tue, 14 Jul 2020 20:04:43 +1000https://danmackinlay.name/notebook/citation_management.htmlBibliographic database Pandoc-citeproc BibTeX/BibLaTeX To mention To avoid docutils citations The genealogy of evidence is important and there are many important ideas about how we could track it, especially with advances in technology; However, this page is not about that propagation of certainty, but rather the shabby proxy, citations in actually-existing academic publishing.
In particular, here I answer for myself: How can I get my journal-ready citations in the 19th century-style format required by my journals with the greatest possible degree of modern possible convenience?Dunning-Kruger theory of mind
https://danmackinlay.name/notebook/dunning_kruger_theory_of_mind.html
Tue, 14 Jul 2020 07:33:53 +1000https://danmackinlay.name/notebook/dunning_kruger_theory_of_mind.htmlOn evaluating others On erroneously thinking my experience is universal On understanding how others think Mitigating References On our known-terrible ability to know our terrible inabilities to know about others. Our constant and reliable ability to 1) think we know more about others than we do and 2) never notice it. Manifesting in, e.g. pluralistic ignorance, the out-group homogeneity effect, idiosyncratic rater effect, many other biases I do not yet know names for.Research discovery
https://danmackinlay.name/notebook/research_discovery.html
Fri, 26 Jun 2020 09:27:43 +1000https://danmackinlay.name/notebook/research_discovery.htmlReading groups and co-learning Paper analysis/annotation Finding copies Recommender systems for academics are hard and in particular, I suspect they are harder than normal because definitionally the content should be new and hard to relate to existing stuff. Indeed, finding connections is a publishable result in itself.
Complicated interaction with systems of peer review. Could a normal recommender system such as canopy be made to work for academics?Soft methodology of science
https://danmackinlay.name/notebook/soft_methodology_of_science.html
Thu, 11 Jun 2020 11:15:12 +1000https://danmackinlay.name/notebook/soft_methodology_of_science.htmlMiscellany of career tips How precious is my idea? Sidling up to the truth. Disruption by field outsiders References The course of science
In which I collect tips from esteemed and eminent minds about how to go about pro-actively discovering stuff. More meta-tips than detailed agendas of discovery.
Miscellany of career tips A current meta-question. One starting point: John Schulman’s Opinionated Guide to ML Research, which discusses stuff like this:Science, history and philosophy thereof
https://danmackinlay.name/notebook/hps.html
Fri, 05 Jun 2020 09:34:20 +1000https://danmackinlay.name/notebook/hps.htmlWhat is science? References I do not myself have a much to say about the philosophy of science as such. I read a lot of Lakatos that one time.
Mostly I am interested in a kind of qualitative mechanism design musing as it pertains to designing better peer-review.
What is science? Not really in that vein, check out amusing curmudgeon: DC Stove, Popper and after: Four modern irrationalists.Knowledge geometry
https://danmackinlay.name/notebook/knowledge_topology.html
Fri, 22 May 2020 10:46:01 +1000https://danmackinlay.name/notebook/knowledge_topology.htmlWhat is the shape of collected human knowledge? To Investigate, possibly related Topic modelling in text databases Artificial chemistry Related links References See also:
Innovation Is a material basis for technology plus a knowledge topology equal to a model of technology? I suspect not - surely there are emergent effects. But there must be a relationship. Spaces of strings String dynamics Related question: What is the shape of the vocabulary of communicating people?Academic writing workflow
https://danmackinlay.name/notebook/academic_writing_workflow.html
Mon, 18 May 2020 10:05:47 +1000https://danmackinlay.name/notebook/academic_writing_workflow.htmlScience blog Writing papers Guides Minimally painful LaTeX collaboration Emailing word files between collaborators “Documenting my academic writing workflow and how to improve it”, or, “How I learned to stop worrying and love text files”.
Science blog Now under academic blogging workflow.
Writing papers Writing papers, especially collaboratively, is a a constant pain point. You might want to try a scientific notebook such as jupyter, knitr etc, which will generate the requisite diagrams etc.Learning with conservation laws, invariances and symmetries
https://danmackinlay.name/notebook/learning_with_conservation_laws.html
Fri, 01 May 2020 09:53:27 +1000https://danmackinlay.name/notebook/learning_with_conservation_laws.htmlReferences Failure of conservation of mass at system boundaries is a common problem in models with nonparametric likelihood
Learning in complicated systems where we know that there is a conservation law in effect. Or, more advanced, learning a conservation law that we did not know was in effect. As seen in especially ML for physics. This is not AFAIK a particular challenge in traditional parametric statistics where we can impose conservation laws on a problem through the likelihood, but nonparametrics models, or models with overparameterisation such as neural nets this can get fiddly.Academic publishing
https://danmackinlay.name/notebook/academic_publishing.html
Thu, 16 Apr 2020 07:56:24 +1000https://danmackinlay.name/notebook/academic_publishing.htmlEconomics of publishing Realpolitik of journals h-Index JIF SJR Tools Prescriptive The new universal libraries Copyright activism Open access References Some notes to the connection between reproducibility, scholarly discovery, intellectual property peer-review, academic business models and such.
Economics of publishing Cameron Neylon runs a cottage industry producing pragmatic publishing critique from an institutional economics perspective:
e.g. The Marginal Costs of Article Publishing or A Journal is a Club:Free will
https://danmackinlay.name/notebook/free_will.html
Thu, 02 Apr 2020 11:14:45 +1100https://danmackinlay.name/notebook/free_will.htmlThe unsatisfying semantic debate that few feel the need to have the correct vocabulary for, but many feel the need to have opinions on. If you are a philosopher of these things, you can move on. I have nothing to add for you. Me, I don’t know what “free will” is, but I know what it isn’t when I see it.
I had a long argument with a drunk gentleman over dinner the other night.Mechanism design for reputation systems
https://danmackinlay.name/notebook/reputation_systems.html
Mon, 02 Mar 2020 08:27:01 +1100https://danmackinlay.name/notebook/reputation_systems.htmlBulletin-board style China-style Iterative reputation References Reputation systems are systems to work out how reliable a bet someone is by the rankings of society. We seem to have some status-seeking which makes us prone to such systems, and the modern technocratic versions are a potential source of mechanisms for collective accomplishment, and or course, control.
I’m mostly thinking about peer ranking systems here, although of course centrally supplied ranking systems are also reputation systems and do indeed fit in here.Selling uncertainty
https://danmackinlay.name/notebook/selling_uncertainty.html
Tue, 18 Feb 2020 10:42:39 +1100https://danmackinlay.name/notebook/selling_uncertainty.htmlWrit small Writ large References Reza Farazmand:
When it’s close it will be larger
On the strategic inception of uncertainty. Fun keyword here: Agnotology, the study of ignorance.
Writ small Possibly one reason this works is the notorious information avoidance problem. (Case et al. 2005)
The assumption that individuals actively seek information underlies much of psychological theory and communication practice, as well as most models of the information-seeking process.R Shiny
https://danmackinlay.name/notebook/shiny.html
Tue, 11 Feb 2020 09:20:13 +1100https://danmackinlay.name/notebook/shiny.htmlShiny basics Containerized webapps Helpers An interactive data anlaysis webapp generator for R.
Shiny basics The RStudio tutorials is well done, and includes video-flavoured and text-flavoured presentations.
You can deploy public apps to the cloud via integrated service shinyapps.io.
Containerized webapps Containerized apps are a convenient way to deploy code in general and to package up an interactive data analysis in particular. The dominant toolchain here is Docker.Bayes for beginners
https://danmackinlay.name/notebook/bayes_howto.html
Thu, 30 Jan 2020 10:15:48 +1100https://danmackinlay.name/notebook/bayes_howto.htmlPrior choice Learning Workflow Nonparametrics Tools Applied References Even for the most currmudgeonly frequentist it is sometimes refreshing to move your effort from deriving frequentist estimators for intractable models, to using the damn Bayesian ones, which fail in different and interesting ways than you are used to. If it works and you are feeling fancy you might then justify your Bayesian method on frequentist grounds, which washes away the sin.Frequentist consistency of Bayesian methods
https://danmackinlay.name/notebook/bayesian_consistency.html
Sat, 19 Oct 2019 09:48:13 +1100https://danmackinlay.name/notebook/bayesian_consistency.htmlReferences You want to use some tasty tool, such as a hierarchical model without anyone getting cross at you for apostasy by doing it in the wrong discipline? Why not use whatever estimator works, and then show that it works on both frequentist and Bayesian grounds?
Shalizi’s overview
There is a basic result here, due to Doob, which essentially says that the Bayesian learner is consistent, except on a set of data of prior probability zero.M-open, M-complete, M-closed
https://danmackinlay.name/notebook/m_open.html
Fri, 18 Oct 2019 22:29:30 +1100https://danmackinlay.name/notebook/m_open.htmlReferences I encountered this concept and thought it was nifty but have not time to do anything other than note it here.
(Le and Clarke 2017) summarise
For the sake of completeness, we recall that Bernardo and Smith (Bernardo and Smith 2000) define M-closed problems as those for which a true model can be identified and written down but is one amongst finitely many models from which an analyst has to choose.