Additional_NetREm_Parameters.md

Additional Parameters for NetREm

Entire usage of the NetREm main function netrem()

NetREm fits a Network-constrained Lasso regression machine learning model with user-provided weights for the prior network. Here, netrem is the main function with the following usage:

netrem(
        edge_list,
        beta_net = 1,
        alpha_lasso = 0.01,
        default_edge_weight = 0.01,
        edge_vals_for_d = True,
        w_transform_for_d = "none",
        degree_threshold = 0.5,
        gene_expression_nodes = [],
        overlapped_nodes_only = False,
        standardize_X = True,
        standardize_y = True,
        center_y = False,
        y_intercept = False,
        view_network = False,
        model_type = "Lasso",
        lasso_selection = "cyclic",
        all_pos_coefs = False,
        tolerance = 1e-4,
        maxit = 10000,
        num_jobs = -1,
        num_cv_folds = 5,
        lassocv_eps = 1e-3,
        lassocv_n_alphas = 100,
        lassocv_alphas = None,
        verbose = False,
        hide_warnings = True
)

The above lists out all of the parameters input into our netrem function. We detail the additional parameters after lasso_selection below. Several of these parameters are based on utilizing Python's scikit-learn library. The main inputs are detailed on the NetREm Home Page.

Additional Inputs:

Parameter	Definition
all_pos_coefs	boolean, default = 'False' Please note that this is the positive parameter found in the Lasso and LassoCV classes in sklearn. • If all_pos_coefs = True, the model will be restricted to be fit with all regression coefficients as positive. • If all_pos_coefs = False, the model will be fit with no restrictions on regression coefficients (can be positive or negative).
tolerance	float, default = 1e-4 The tolerance sklearn would use for optimizing the NetREm model. (This is known as tol in by Python's sklearn). If the updates to the optiimzation are smaller than tolerance, then the optimization code will check the dual gap for optimizality and contine the optimization until that dual gap is smaller than tolerance.
maxit	int, default = 10000 Please note that this is the max_iter parameter found in the Lasso and LassoCV classes in sklearn. This is the maximum number of iterations that NetREm will perform.
verbose	boolean, default = False If True, this will print additional lines and details of steps and initial results to the console when running netrem.
hide_warnings	boolean, default = True If True, this will display warnings generated by Python when running netrem.

• Parameters if model_type = LassoCV are derived from the LassoCV class in sklearn:

Parameter	Definition
lassocv_eps	float, default = 1e-3 This corresponds to the eps epsilon parameter in LassoCV. It is the length of the path. Here, lassocv_eps = 1e-3 means that alpha_min / alpha_max = 1e-3.
lassocv_n_alphas	int, default = 100 This corresponds to the n_alphas parameter in LassoCV. This is the number of alphas along the Lasso regularization path.
lassocv_alphas	array-like, default = None This corresponds to the alphas parameter in LassoCV. List of alphas where the models to be computed. If None then the alphas are set automatically.
num_cv_folds	float, default = 5 By default, sklearn cross-validation is used. This specifies the number of folds for splitting the training data when fitting the NetREm model.
num_jobs	int, default = -1 Number of computational jobs to run in parallel, which is used to determine burden, efficiency, and load (resource management). ⌛ None means 1 unless in a joblib.parallel_backend context. -1 means using all of the processors. This is similar to the n_jobs parameter in Python's sklearn.