dev/developingNotes.md

In alexholcombe/mixRSVP: Mixture Model Temporal Errors In RSVP Data

To-do

• I cast warnings as a character to fix a problem but side effect is getting these messages "Unequal factor levels: coercing to characterbinding character and factor vector, coercing into character vectorbinding character and factor vector"

to-do

• [1] "Got NaN as a result of calling pnorm with: 1.5 -0.227865514090563 -4.05e-05" Why is pnorm being called with a negative sigma?? Happens on line 51 or likelihood.R Need to find what is setting sigma to negative value.

• Create simplest possible vignette, to analyze just one condition. Basically filter data and then call analyzeOneCondition

• Does the namespace for annotate() , ggplot2::annotate somehow get screwed up so you have to specify ggplot2::annotate?

• to avoid writing the text one time for each data point, would like to cut to one trial per condition, but can't do that without

  #knowing the grouping variable for the facets #how do I keep one x from each condition? This is basically like stat_summary #See https://stackoverflow.com/questions/15720545/use-stat-summary-to-annotate-plot-with-number-of-observations

#No, unfortunately stat_summary only gets the numbers, this is the way to do it https://stackoverflow.com/a/46791112/302378 Have to create separate dataset and geom_text call to add the annotations

• Add set.seed(1) # Reproducibility to everything

• Add inferential stats to analyze_experiment.Rmd vignette

• Pesky fitting errors. Error in grad.default(ufn, ans$par, ...) : function returns NA at 1.4434927002511e-050.000135401581392880.000100001 distance from x. I have started a test to investigate this in test-fit_model.R

• Add for the nReplicates parameter roxygen a link to the notes on how many fits are needed for stability

• add a link somewhere, somehow to BackwardsLtrs (Google Drive) E2/compareMATLABtoR.html

Stats

For comparing models, for the AB case of having two episodes, Pat used Bayesian Information Criteria (BIC) for the single- and dual-episode models, based on the combined T1 and T2 distribution, to determine when there is evidence of a second attentional episode. in AB_compare_models.m

For comparing guessing only to mixture, Pat did:

% % Test for a significant difference in log likelihoods % [h,pValue,stat,cValue] = lratiotest(-minNegLogLikelihood,-uniformNegLogLikelihood,nFreeParameters,pCrit);

The mixture model has 3 degrees of freedom I think (e, u, sigma) and the guessing distrbiutino zero.

• Need to add Pat's quantitative criterion (or similar) to exclude subjects with low efficacy. Have got "excluded field" from his MATLAB fits in backwards2_E1 that I can compare my test with.

Questions

insanely, format.p cannot be found by annotate_fit call from top environment even when it's within the function defined

• Would be ridiculous to use ggplot2:: everytime invoked that ,but is that the way to get rid of "plot_hist_with_fit: no visible binding for global variable ‘SPE’" ?

Testing

Use RStudio->Build->Test or

Damn you can't see the vignettes unless you install the proper package from tgz etc or https://stackoverflow.com/questions/33614660/knitr-rmd-vignettes-do-not-appear-with-vignette?rq=1

This can be investigated in test_analyzeOneCondition's test "can combine results of analyzeOneConditionDf into dataframe"

analyzeOneConditionDF turned warnings into a string.

Column warnings can't be converted from character to list. Fixed

inside calc_curves_dataframe in test-calc_curves_dataframe in bind_rows_(x, .id) : Column warnings can't be converted from character to list

Still getting this error e.g. in inspectHistograms.Rmd in jackie project. Do I have the latest version? Because what is the relationship to the message given by analyze_experiment.Rmd: " Unequal factor levels: coercing to characterbinding ". I dimly remember that "binding character and factor vector, coercing into character vector"

"call" in simplewarning sometimes has two fucking parts!

Vignettes

For a long time the vignettes were lost no matter which RStudio build thing I ran. Until I did devtools::build_vignettes() devtools::load_all(".")

Then I build source package in case that's needed to install from GitHub, but it doesn't work. Turns out a further complication is that build_vignettes is no longer an option in install_github, instead you have to do: devtools::install_github('alexholcombe/mixRSVP', build = TRUE, build_opts = c("--no-resave-data", "--no-manual")) This does not include the option of not building the vignettes, so it results in building the vignettes.

For adding an image for a vignette, it should also be in the vignette source directory

alexholcombe/mixRSVP documentation built on June 7, 2019, 3:50 p.m.