model_features: Model features
In pmcharrison/voicer: Analyzing and Generating Chord Voicings

This function fits a sequential conditional logit model that predicts chord voicings from chord features.

1
2
3

model_features(x, features = attr(x, "features"),
  formula = linear_formula(x, features), keep_model = TRUE,
  eval_model = TRUE, perm_int = TRUE, verbose = TRUE)

`x`	A corpus's feature analysis as created by `get_corpus_features`.
`features`	(Character vector) Features to model, corresponding to columns of `x`. By default this is extracted from the metadata of `x`, and corresponds to all computed features.
`formula`	(Formula) The formula used to predict chord voicings. The default is created by `linear_formula`, and corresponds to a simple linear combination of the features specified in the `features` argument.
`keep_model`	(Logical scalar) Determines whether the returned object should include the fitted model from `mclogit`, which can be rather large.
`eval_model`	(Logical scalar) Whether to compute and return model evaluation metrics.
`perm_int`	(Logical scalar) Whether to compute and return permutation-based feature importance metrics (also known as model reliance metrics).
`verbose`	(Logical scalar) Whether to show progress messages during the computation.

An object of class voicer_mod with the following slots:

features - A character vector listing the features used to fit the model.
formula - A character vector corresponding to the formula used to fit the model.
weights - A tibble describing the fitted weights. This tibble has five columns: feature (the name of the feature, or more generally the effect), estimate (the estimate of the corresponding regression weight), std_err (the standard error of the regression weight), z (the z-statistic associated with the regression weight), and p (the p-value associated with the regression weight).
mod - A fitted model as created by mclogit.
eval - Evaluation metrics for the fitted model. This is a list with two components. summary is a tibble listing six model statistics: probability, the mean probability assigned to the observed voicing; accuracy, the mean accuracy when predicting the correct voicing; info_content, the mean log probability of the observed voicing (base 2); num_options, the mean number of candidate voicings for each chord; abs_rank, the mean absolute rank of the chosen voicing within the list of candidates as ranked by assigned probability; pct_rank, the mean percentile rank of the chosen voicing within the list of candidates as ranked by assigned probability. perm_int is a tibble listing permutation importances (also known as model reliances) by feature, with the permutation importance metrics mapping to the columns of summary.