unfoldtoolbox / unfoldsim.jl Goto Github PK

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

1. Would be nice to format a code in Home page in docs.
2. It crashed when I entered 8 at σ in onset = LogNormalOnset(μ = 5, σ = 5.1). Extrema parameters should be tested.
3. Here would be nice to have more spaces in label = "($μ,$σ,$offset,$truncate_upper)". Hard to read
4. DocStrings probably should renamed to API
5. UnfoldSim.predef_eeg - this docstring definitely should be check for grammar and formatting.
6. I solved the issue of _ transforming to italic in my docstrings. You can check it here https://github.com/unfoldtoolbox/UnfoldMakie.jl/blob/main/src/docstring_template.jl on string 48
7. UnfoldSim.simulate_component - 4 spaces should be put before a function in doctrings
8. Each docstring should end with a returned value

We have to profile a run to see why we are relatively slow (~200ms for simulating 30subjects, 100 trials with LMMs). I want to be much faster :)

e.g.
generate(SingleSubjectDesign(;) |> x-> RepeatDesign(x,7))

add error message

Before 1.0 we should rename a few things:

• simulate and generate are used in different places, not clear why not same name for everything. Makes sense to me, you can simulate a design, simulate an EEG, simulate noise
• rand_onsets actually delivers distances between onsets, generate(rng,onset,...) generates the onsets

Let's collect as we go

Currently not clear how to best do something like
A -> B -> C as three different events, or even worse, B follows A in 80% of cases, etc.

• module for the simulating underlying EEG signal data.
  • per event: one set of coefficients, able to simulate random effect structure subject + item
  • from mathematical functions - exponential functions etc
  • from filters - hanning window etc

I will elaborate them later in the thread as we proceed further.

• Readme
• Check dependencies
• Add Compat entries
• Author => Luis Lips, Benedikt Ehinger, Judith Schepers,
• Setup Continuous Integration (maybe redoing the package is easier as we can use PkgTemplate.jl?
• We need unittests
• We need documentation

Here we collect ideas that we might want to simulate and look at

• Timefrequency response to overlapping signals (with @olafdimigen)
• discrete vs. continuous jitter
• duration effects
• length of estimation window
• number of parameters/predictors/splines/events (@ReneSkukies)
• show that bias in e.g. saccade-amplitude leads to spurios decoding
• have a second look at sequence effects (described in Gert et al 2022)
• simulate patterns for erpimages to autodetect interesting covariates (@vladdez )
• overlap via source-activation, then do source inversion, then check if Unfold still has to do something

Feel free to add your own ideas!

• Simulate outliers
• Option to save simulated data in typical EEG format(s)
• Simulate ECG signal

right now we can only do pseudo-continuous via discretizing [1,2,3,4,5] - would be good to have real continuous sampling ability

Simulate events and EEG data based on

• StatsBase @formula
• Trials, Times, channels (more flexibility, in terms of choosing covariate)

https://github.com/unfoldtoolbox/UnfoldSim.jl/blob/21e46a4c7ef1b586ed324e24d3bb5552c574fdf4/src/component.jl#L189C1-L189C51

split this line in a new subfunction, to allow extraction of simulated BLUPs

Luis, please dont do any of those - you have your project/thesis! but if you have thoughts happy to hear them :)

• add some basic basis functions (ERP-style, P300, N400, N170, fMRI, PupilDilation etc.)
• provide Docs to these basis functions with pretty example
• add realistic EEG Noise - maybe based on Artifacts.jl or DataDeps.jl so not everyone needs to download it
• provide a simplistic doc HowTo to run a power analysis (out of scope for this package? dunno)

• if Condition/level is an int, I get an error later
• if #betas or #thetas doesnt fit, I get a hard to parse error
• convert generates a "cond" not a "stimType" column

@all-contributors please add @behinger for bug,code,doc,ideas,infra,maintenance,review,test,tutorial

function simulate(args...; kwargs...)
    @warn "No random generator defined, used the default (`Random.MersenneTwister(1)`) with a fixed seed. This will always return the same results and the user is strongly encouraged to provide their own random generator!"
    simulate(MersenneTwister(1), args...; kwargs...)
end

on second thought, that wasnt our best idea. As soon as you misspecify the simulate input in any way (e.g. NoNoise instead of NoNoise()) it runs into an infinite loop

When creating a MultiSubjectDesign without giving arguments for subjects_between, items_between and both_within (their default is nothing), the generate function gives a MethodError as in the following example:

The problem originates from the following line in the generate function:

Suggestions:

• Either rewrite the function to make it possible to generate designs without any conditions
• Or handle this case with an appropriate error + message.

Issue to call the JuliaBot

@JuliaRegistrator register

Release notes:

• Initial Release. Toolbox to simulate event related time series. Focus on EEG

• realistic EEG Noise - maybe based on Artifacts.jl or DataDeps.jl so not everyone needs to download it. Currently discussing with @ReneSkukies how to do this
• Autocorrelated noise - @jschepers there was a paper we cite that had a nice noise-generator, do you remember?
• Exponential Noise - already implemented, but we should cap the exponential function at some point and apply it in batches or so
• Noise á la carte - implement the approach from Richard Gao on generating arbitrary complex signal-stubs from short segments (or generate data with their implementation and use Artifacts.jl to serve it)

the bug I have seen in @vladdez notebook with the event_order_function appeared again. I have to test and drill it down further, but it appeared now with a sequence+repetition design model. Unclear if without these it would appear too

without shuffle:

with shuffle:

clearly the conditions are not separated correctly.

There is a code example on the main page: https://github.com/unfoldtoolbox/UnfoldSim.jl

data,evts = UnfoldSim.predef_eeg(;n_trials=20,noiselevel=0.8)
I got this error:

But everything works after:

data, evts = UnfoldSim.predef_eeg(;n_repeats=20, noiselevel=0.8, return_epoched=true)