Setup

using Unfold
using UnfoldMakie, CairoMakie
using UnfoldSim
using UnfoldDecode
using DataFrames
using Statistics

Quick start of Back-To-Back (b2b)

Data generation

Simulation and data collection

Generate single channel data via UnfoldSim.jl

dat, evts = UnfoldSim.predef_eeg(; noiselevel = 0.1, return_epoched = true);

dat is a time x repetition Matrix, evts is a DataFrame`with independent variables / features to explain the data

Dimension expansion

Repeat the dat 20 times, representing 20 channels. In the future we will replace this with a direct multi-channel simulation.

dat_3d = permutedims(repeat(dat, 1, 1, 20), [3 1 2]);

Channels shouldnt be identical, so we add some noise.

dat_3d .+= 0.1 * rand(size(dat_3d)...);

one could specify the specific solvers for G and H by passing the solver_G and solver_H parameters to the solver_b2b function. Implemented solvers are ridge, lasso, lsq, svm, but other solvers from MLJ.jl can be used as well.

Generate the formula

We want to decode condition, but simultaneously control for the effect of continuous.

f = @formula 0 ~ 1 + condition + continuous
time = range(0, 0.44, step = 1 / 100)
designDict = [Any => (f, time)]

m = Unfold.fit(UnfoldModel, designDict, evts, dat_3d; solver = b2b_solver);

results = coeftable(m);
results.estimate = abs.(results.estimate); ## back2back has no sign
results = results[results.coefname.!="(Intercept)", :] ## the intercept in b2b is hard to interpret

90×7 DataFrame

Row	channel	coefname	estimate	eventname	group	stderror	time
	Int64	String	Float64	DataType	Nothing	Nothing	Float64
1	1	condition: face	0.00201176	Any			0.0
2	1	condition: face	0.00364378	Any			0.01
3	1	condition: face	0.00520939	Any			0.02
4	1	condition: face	0.00501611	Any			0.03
5	1	condition: face	0.00143283	Any			0.04
6	1	condition: face	0.00233638	Any			0.05
7	1	condition: face	0.00922447	Any			0.06
8	1	condition: face	0.00927352	Any			0.07
9	1	condition: face	0.00962683	Any			0.08
10	1	condition: face	0.00564038	Any			0.09
11	1	condition: face	0.0579659	Any			0.1
12	1	condition: face	0.194406	Any			0.11
13	1	condition: face	0.5796	Any			0.12
14	1	condition: face	0.307346	Any			0.13
15	1	condition: face	0.270874	Any			0.14
16	1	condition: face	0.270424	Any			0.15
17	1	condition: face	0.272506	Any			0.16
18	1	condition: face	0.282857	Any			0.17
19	1	condition: face	0.31145	Any			0.18
20	1	condition: face	0.390778	Any			0.19
21	1	condition: face	0.0365178	Any			0.2
22	1	condition: face	0.0222499	Any			0.21
23	1	condition: face	0.00707891	Any			0.22
24	1	condition: face	0.00644083	Any			0.23
25	1	condition: face	0.00647094	Any			0.24
26	1	condition: face	0.00673642	Any			0.25
27	1	condition: face	0.00644951	Any			0.26
28	1	condition: face	0.00599373	Any			0.27
29	1	condition: face	0.00647065	Any			0.28
30	1	condition: face	0.00632556	Any			0.29
31	1	condition: face	0.00664532	Any			0.3
32	1	condition: face	0.00664224	Any			0.31
33	1	condition: face	0.00676942	Any			0.32
34	1	condition: face	0.00704748	Any			0.33
35	1	condition: face	0.00704868	Any			0.34
36	1	condition: face	0.00689335	Any			0.35
37	1	condition: face	0.00695012	Any			0.36
38	1	condition: face	0.00680078	Any			0.37
39	1	condition: face	0.00590307	Any			0.38
40	1	condition: face	0.00730209	Any			0.39
41	1	condition: face	0.00706259	Any			0.4
42	1	condition: face	0.00870782	Any			0.41
43	1	condition: face	0.012643	Any			0.42
44	1	condition: face	0.0253075	Any			0.43
45	1	condition: face	0.0162129	Any			0.44
46	1	continuous	0.000270012	Any			0.0
47	1	continuous	6.95933e-6	Any			0.01
48	1	continuous	0.000200561	Any			0.02
49	1	continuous	0.000127116	Any			0.03
50	1	continuous	0.000350511	Any			0.04
51	1	continuous	7.50886e-5	Any			0.05
52	1	continuous	0.000289418	Any			0.06
53	1	continuous	1.35741e-5	Any			0.07
54	1	continuous	5.88929e-5	Any			0.08
55	1	continuous	6.73665e-7	Any			0.09
56	1	continuous	6.22823e-7	Any			0.1
57	1	continuous	2.80815e-5	Any			0.11
58	1	continuous	7.95363e-5	Any			0.12
59	1	continuous	0.000250677	Any			0.13
60	1	continuous	0.000129402	Any			0.14
61	1	continuous	0.00012792	Any			0.15
62	1	continuous	3.74312e-6	Any			0.16
63	1	continuous	0.000914131	Any			0.17
64	1	continuous	0.00809811	Any			0.18
65	1	continuous	0.0939346	Any			0.19
66	1	continuous	0.857561	Any			0.2
67	1	continuous	0.358082	Any			0.21
68	1	continuous	0.273539	Any			0.22
69	1	continuous	0.278744	Any			0.23
70	1	continuous	0.28211	Any			0.24
71	1	continuous	0.287228	Any			0.25
72	1	continuous	0.286356	Any			0.26
73	1	continuous	0.279863	Any			0.27
74	1	continuous	0.281136	Any			0.28
75	1	continuous	0.284399	Any			0.29
76	1	continuous	0.281916	Any			0.3
77	1	continuous	0.286074	Any			0.31
78	1	continuous	0.284982	Any			0.32
79	1	continuous	0.281391	Any			0.33
80	1	continuous	0.282507	Any			0.34
81	1	continuous	0.280331	Any			0.35
82	1	continuous	0.27924	Any			0.36
83	1	continuous	0.278484	Any			0.37
84	1	continuous	0.275909	Any			0.38
85	1	continuous	0.273713	Any			0.39
86	1	continuous	0.2579	Any			0.4
87	1	continuous	0.245107	Any			0.41
88	1	continuous	0.192979	Any			0.42
89	1	continuous	0.0516782	Any			0.43
90	1	continuous	3.68542e-5	Any			0.44

Plotting

plot_erp(results; axis = (xlabel = "Time [s]", ylabel = "Performance"))

We can see from the graph that b2b solver identifies regions where the signal can be decoded, taking into account the continuous feature

This page was generated using Literate.jl.

Setup

Quick start of Back-To-Back (b2b)

Data generation

Simulation and data collection

Dimension expansion

Modeling

Solver selection

Generate the formula

Plotting