dfr_sgl {dfr}R Documentation

Fit a DFR-SGL model.

Description

Sparse-group lasso (SGL) with DFR main fitting function. Supports both linear and logistic regression, both with dense and sparse matrix implementations.

Usage

dfr_sgl(
  X,
  y,
  groups,
  type = "linear",
  lambda = "path",
  alpha = 0.95,
  max_iter = 5000,
  backtracking = 0.7,
  max_iter_backtracking = 100,
  tol = 1e-05,
  standardise = "l2",
  intercept = TRUE,
  path_length = 20,
  min_frac = 0.05,
  screen = TRUE,
  verbose = FALSE
)

Arguments

X

Input matrix of dimensions n \times p. Can be a sparse matrix (using class "sparseMatrix" from the Matrix package).

y

Output vector of dimension n. For type="linear" should be continuous and for type="logistic" should be a binary variable.

groups

A grouping structure for the input data. Should take the form of a vector of group indices.

type

The type of regression to perform. Supported values are: "linear" and "logistic".

lambda

The regularisation parameter. Defines the level of sparsity in the model. A higher value leads to sparser models:

  • "path" computes a path of regularisation parameters of length "path_length". The path will begin just above the value at which the first predictor enters the model and will terminate at the value determined by "min_frac".

  • User-specified single value or sequence. Internal scaling is applied based on the type of standardisation. The returned "lambda" value will be the original unscaled value(s).

alpha

The value of \alpha, which defines the convex balance between the lasso and group lasso. Must be between 0 and 1. Recommended value is 0.95.

max_iter

Maximum number of ATOS iterations to perform.

backtracking

The backtracking parameter, \tau, as defined in Pedregosa and Gidel (2018).

max_iter_backtracking

Maximum number of backtracking line search iterations to perform per global iteration.

tol

Convergence tolerance for the stopping criteria.

standardise

Type of standardisation to perform on X:

  • "l2" standardises the input data to have \ell_2 norms of one. When using this "lambda" is scaled internally by 1/\sqrt{n}.

  • "l1" standardises the input data to have \ell_1 norms of one. When using this "lambda" is scaled internally by 1/n.

  • "sd" standardises the input data to have standard deviation of one.

  • "none" no standardisation applied.

intercept

Logical flag for whether to fit an intercept.

path_length

The number of \lambda values to fit the model for. If "lambda" is user-specified, this is ignored.

min_frac

Smallest value of \lambda as a fraction of the maximum value. That is, the final \lambda will be "min_frac" of the first \lambda value.

screen

Logical flag for whether to apply the DFR screening rules (see Feser and Evangelou (2024)).

verbose

Logical flag for whether to print fitting information.

Details

dfr_sgl() fits a DFR-SGL model (Feser and Evangelou (2024)) using Adaptive Three Operator Splitting (ATOS) (Pedregosa and Gidel (2018)). It solves the convex optimisation problem given by (Simon et al. (2013))

\frac{1}{2n} f(b ; y, \mathbf{X}) + \lambda \alpha \sum_{i=1}^{p} |b_i| + \lambda (1-\alpha)\sum_{g=1}^{m} \sqrt{p_g} \|b^{(g)}\|_2,

where f(\cdot) is the loss function and p_g are the group sizes. In the case of the linear model, the loss function is given by the mean-squared error loss:

f(b; y, \mathbf{X}) = \left\|y-\mathbf{X}b \right\|_2^2.

In the logistic model, the loss function is given by

f(b;y,\mathbf{X})=-1/n \log(\mathcal{L}(b; y, \mathbf{X})).

where the log-likelihood is given by

\mathcal{L}(b; y, \mathbf{X}) = \sum_{i=1}^{n}\left\{y_i b^\intercal x_i - \log(1+\exp(b^\intercal x_i)) \right\}.

SGL can be seen to be a convex combination of the lasso and group lasso, balanced through alpha, such that it reduces to the lasso for alpha = 1 and to the group lasso for alpha = 0. By applying both the lasso and group lasso norms, SGL shrinks inactive groups to zero, as well as inactive variables in active groups. DFR uses the dual norm (the \epsilon-norm) and the KKT conditions to discard features at \lambda_k that would have been inactive at \lambda_{k+1}. It applies two layers of screening, so that it first screens out any groups that satisfy

\|\nabla_g f(\hat{\beta}(\lambda_{k}))\|_{\epsilon_g} \leq \tau_g(2\lambda_{k+1} - \lambda_k)

and then screens out any variables that satisfy

|\nabla_i f(\hat{\beta}(\lambda_{k}))| \leq \alpha (2\lambda_{k+1} - \lambda_k)

leading to effective input dimensionality reduction. See Feser and Evangelou (2024) for full details.

Value

A list containing:

beta

The fitted values from the regression. Taken to be the more stable fit between x and z, which is usually the former. A filter is applied to remove very small values, where ATOS has not been able to shrink exactly to zero. Check this against x and z.

group_effects

The group values from the regression. Taken by applying the \ell_2 norm within each group on beta.

selected_var

A list containing the indicies of the active/selected variables for each "lambda" value. Index 1 corresponds to the first column in X.

selected_grp

A list containing the indicies of the active/selected groups for each "lambda" value. Index 1 corresponds to the first group entry in the groups vector. You can see the group order by running unique(groups).

num_it

Number of iterations performed. If convergence is not reached, this will be max_iter.

success

Logical flag indicating whether ATOS converged, according to tol.

certificate

Final value of convergence criteria.

x

The solution to the original problem (see Pedregosa and Gidel (2018)).

u

The solution to the dual problem (see Pedregosa and Gidel (2018)).

z

The updated values from applying the first proximal operator (see Pedregosa and Gidel (2018)).

screen_set_var

List of variables that were kept after screening step for each "lambda" value. (see Feser and Evangelou (2024)).

screen_set_grp

List of groups that were kept after screening step for each "lambda" value. (see Feser and Evangelou (2024)).

epsilon_set_var

List of variables that were used for fitting after screening for each "lambda" value. (see Feser and Evangelou (2024)).

epsilon_set_grp

List of groups that were used for fitting after screening for each "lambda" value. (see Feser and Evangelou (2024)).

kkt_violations_var

List of variables that violated the KKT conditions each "lambda" value. (see Feser and Evangelou (2024)).

kkt_violations_grp

List of groups that violated the KKT conditions each "lambda" value. (see Feser and Evangelou (2024)).

screen

Logical flag indicating whether screening was performed.

type

Indicates which type of regression was performed.

intercept

Logical flag indicating whether an intercept was fit.

lambda

Value(s) of \lambda used to fit the model.

References

Feser, F., Evangelou, M. (2024). Dual feature reduction for the sparse-group lasso and its adaptive variant, https://arxiv.org/abs/2405.17094

Pedregosa, F., Gidel, G. (2018). Adaptive Three Operator Splitting, https://proceedings.mlr.press/v80/pedregosa18a.html

Simon, N., Friedman, J., Hastie, T., Tibshirani, R. (2013). A Sparse-Group Lasso, doi:10.1080/10618600.2012.681250

See Also

Other SGL-methods: dfr_adap_sgl(), dfr_adap_sgl.cv(), dfr_sgl.cv(), plot.sgl(), predict.sgl(), print.sgl()

Examples

# specify a grouping structure
groups = c(1,1,1,2,2,3,3,3,4,4)
# generate data
data = sgs::gen_toy_data(p=10, n=5, groups = groups, seed_id=3,group_sparsity=1)
# run DFR-SGL 
model = dfr_sgl(X = data$X, y = data$y, groups = groups, type="linear", path_length = 5, 
alpha=0.95, standardise = "l2", intercept = TRUE, verbose=FALSE)

[Package dfr version 0.1.5 Index]