Changelog • neurogam

neurogam 0.0.4

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), ...).
Implementing post-processing methods (e.g., print(), summary(), plot()) for 2D temporal models.
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).

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.
Allowing to save the fitted brms model in testing_through_time() via the save argument.
Re-factoring the check_residual_autocorrelation() function to use residuals.brmsfit() and returns credible intervals.

Fixing the specification of AR terms in make_bgam_formula() (now works for both participant- and group-level models).
The check_residual_autocorrelation() function now works when multilevel = "group".

Implementing the recommend_k() function, running model comparison with varying basis dimension (k).
Implementing a ppc() method for cluster_results objects. (#2)
Adding support for modelling auto-correlation via testing_through_time(include_ar_term == TRUE).
Implementing the check_residual_autocorrelation() function to visualise auto-correlation of the residuals.
New plot_sensors() and plot_eeg() functions to visualise EEG sensors grid, EEG raw data, and GAM predictions.
Adding support for returning participant-level onsets in testing_through_time(), find_clusters(), and related methods such as summary() and plot(). (#4)
Adding support for binomial() responses. (#7)
Adding support for continuous predictors in testing_through_time(). (#1)
Allowing to pass a precomputed outcome SD to testing_through_time().
Adding the use_se argument in testing_through_time() and make_bgam_formula().

Improved documentation for print() and summary() methods.
Removed the multilevel = "full" option (too slow).
Now returns clusters with both positive and negative signs in find_clusters().
Allowing to manually specify fill limits in plot_eeg().
Allowing ggplot2 theme to be modified in plotting functions.
Providing a more informative error message after internal data summary when outcome_sd contains NAs.

Fixing aberrant PPCs in the ppc() method (and simplifying arguments with ppc_type = c("group", "participant")).
Allowing “negative” clusters in cluster_results print(), summary(), and plot() methods.
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.

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.

Fixing group-level posterior predictions when multilevel is “full” or “summary”.