Ensure cluster detection at the participant level use the participant-specific baseline-derived ROPE (cf. utils.R).
Other changes
Fixing a code error and using more efficient Duchon splines in the spatio-temporal vignette, thanks to Jiahui Fan.
neurogam 0.1.0
Preparing for initial CRAN submission (updating and reorganising the vignettes, cleaning the documentation, etc).
neurogam 0.0.5
New features
Implementing spatio-temporal BGAMMs for modelling EEG data in both space and time. (#8)
Unified API for processing results.
Re-structuring the find_clusters() approach (now work with any number of dimensions and accepts input adjacency matrices).
Adding a baseline_window argument to testing_through_time() used to compute the ROPE (i.e., maximum absolute value of the credible_interval% credible interval during this baseline period).
This baseline approach leads to the new ROPE default (i.e., odds_mode = "directional_rope") for 1D temporal data.
When including varying smooths in make_bgam_formula(), the default is now to exclude varying intercept and slope.
When including varying smooths in make_bgam_formula(), the default is now to use a k-value of floor(kvalue/2) for the varying (by-participant) smooths.
Bug fixes
Fixing the stan_model_args issue when using the rstan backend, thanks to Maël Delem. (#10)
Fixing the se() value (should be SEM instead of SD).
neurogam 0.0.4
New features
Implementing 2D temporal BGAMs, which can be useful for modelling cross-temporal decoding generalisation matrices. Such models can be fitted by defining two time columns (i.e., training and testing times) in testing_through_time(). For instance, testing_through_time(..., time_id = c(train_time, test_time), ...).
Adding a previous_model argument to testing_through_time(). This allows to pass an existing neurogam model (previously fitted with testing_through_time()), which may be useful for exploring the effects of the threshold parameters (this avoids re-fitting the model).
Other changes
Including the timegen_data (2D temporal data).
Including a new vignette on 2D temporal models using the timegen_data data.
Including a new vignette on checking and modelling residual auto-correlation (for 1D temporal data).
Allowing to use within-chain parallelisation in testing_through_time() via the threads argument.
Fixing error in testing_through_time() when multilevel = "summary" and include_ar_term == TRUE.
Fixing the varying_smooth_term error in make_bgam_formula() when predictor_id = NA.
neurogam 0.0.2
New features
Allowing 3 different models to be fitted: full GAMM, GAMM with summary statistics (recommended), or group-level GAM.
Adding further support for presence or absence of predictor (e.g., group, condition). When predictor_id = NA, neurogam now tests signal against 0 through time.
Implementing print() and summary() methods for cluster_results objects.
Improved plotting: now plotting the GAM predictions with raw data and improved clusters aesthetics.
Other changes
Improved functions documentation and new package website.
Factoring posterior odds computation within internal functions.
Bug fixes
Fixing group-level posterior predictions when multilevel is “full” or “summary”.