Builds a generalized low rank decomposition of an H2O data frame
Usage
h2o.glrm(
training_frame,
cols = NULL,
model_id = NULL,
validation_frame = NULL,
ignore_const_cols = TRUE,
score_each_iteration = FALSE,
representation_name = NULL,
loading_name = NULL,
transform = c("NONE", "STANDARDIZE", "NORMALIZE", "DEMEAN", "DESCALE"),
k = 1,
loss = c("Quadratic", "Absolute", "Huber", "Poisson", "Hinge", "Logistic", "Periodic"),
loss_by_col = c("Quadratic", "Absolute", "Huber", "Poisson", "Hinge", "Logistic",
"Periodic", "Categorical", "Ordinal"),
loss_by_col_idx = NULL,
multi_loss = c("Categorical", "Ordinal"),
period = 1,
regularization_x = c("None", "Quadratic", "L2", "L1", "NonNegative", "OneSparse",
"UnitOneSparse", "Simplex"),
regularization_y = c("None", "Quadratic", "L2", "L1", "NonNegative", "OneSparse",
"UnitOneSparse", "Simplex"),
gamma_x = 0,
gamma_y = 0,
max_iterations = 1000,
max_updates = 2000,
init_step_size = 1,
min_step_size = 1e-04,
seed = -1,
init = c("Random", "SVD", "PlusPlus", "User"),
svd_method = c("GramSVD", "Power", "Randomized"),
user_y = NULL,
user_x = NULL,
expand_user_y = TRUE,
impute_original = FALSE,
recover_svd = FALSE,
max_runtime_secs = 0,
export_checkpoints_dir = NULL
)Arguments
- training_frame
Id of the training data frame.
- cols
(Optional) A vector containing the data columns on which k-means operates.
- model_id
Destination id for this model; auto-generated if not specified.
- validation_frame
Id of the validation data frame.
- ignore_const_cols
Logical. Ignore constant columns. Defaults to TRUE.- score_each_iteration
Logical. Whether to score during each iteration of model training. Defaults to FALSE.- representation_name
Frame key to save resulting X
- loading_name
[Deprecated] Use representation_name instead. Frame key to save resulting X.
- transform
Transformation of training data Must be one of: "NONE", "STANDARDIZE", "NORMALIZE", "DEMEAN", "DESCALE". Defaults to NONE.
- k
Rank of matrix approximation Defaults to 1.
- loss
Numeric loss function Must be one of: "Quadratic", "Absolute", "Huber", "Poisson", "Hinge", "Logistic", "Periodic". Defaults to Quadratic.
- loss_by_col
Loss function by column (override) Must be one of: "Quadratic", "Absolute", "Huber", "Poisson", "Hinge", "Logistic", "Periodic", "Categorical", "Ordinal".
- loss_by_col_idx
Loss function by column index (override)
- multi_loss
Categorical loss function Must be one of: "Categorical", "Ordinal". Defaults to Categorical.
- period
Length of period (only used with periodic loss function) Defaults to 1.
- regularization_x
Regularization function for X matrix Must be one of: "None", "Quadratic", "L2", "L1", "NonNegative", "OneSparse", "UnitOneSparse", "Simplex". Defaults to None.
- regularization_y
Regularization function for Y matrix Must be one of: "None", "Quadratic", "L2", "L1", "NonNegative", "OneSparse", "UnitOneSparse", "Simplex". Defaults to None.
- gamma_x
Regularization weight on X matrix Defaults to 0.
- gamma_y
Regularization weight on Y matrix Defaults to 0.
- max_iterations
Maximum number of iterations Defaults to 1000.
- max_updates
Maximum number of updates, defaults to 2*max_iterations Defaults to 2000.
- init_step_size
Initial step size Defaults to 1.
- min_step_size
Minimum step size Defaults to 0.0001.
- seed
Seed for random numbers (affects certain parts of the algo that are stochastic and those might or might not be enabled by default). Defaults to -1 (time-based random number).
- init
Initialization mode Must be one of: "Random", "SVD", "PlusPlus", "User". Defaults to PlusPlus.
- svd_method
Method for computing SVD during initialization (Caution: Randomized is currently experimental and unstable) Must be one of: "GramSVD", "Power", "Randomized". Defaults to Randomized.
- user_y
User-specified initial Y
- user_x
User-specified initial X
- expand_user_y
Logical. Expand categorical columns in user-specified initial Y Defaults to TRUE.- impute_original
Logical. Reconstruct original training data by reversing transform Defaults to FALSE.- recover_svd
Logical. Recover singular values and eigenvectors of XY Defaults to FALSE.- max_runtime_secs
Maximum allowed runtime in seconds for model training. Use 0 to disable. Defaults to 0.
- export_checkpoints_dir
Automatically export generated models to this directory.
Value
an object of class H2ODimReductionModel.
References
M. Udell, C. Horn, R. Zadeh, S. Boyd (2014). Generalized Low Rank Models[https://arxiv.org/abs/1410.0342]. Unpublished manuscript, Stanford Electrical Engineering Department. N. Halko, P.G. Martinsson, J.A. Tropp. Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions[https://arxiv.org/abs/0909.4061]. SIAM Rev., Survey and Review section, Vol. 53, num. 2, pp. 217-288, June 2011.
Examples
if (FALSE) { # \dontrun{
library(h2o)
h2o.init()
australia_path <- system.file("extdata", "australia.csv", package = "h2o")
australia <- h2o.uploadFile(path = australia_path)
h2o.glrm(training_frame = australia, k = 5, loss = "Quadratic", regularization_x = "L1",
gamma_x = 0.5, gamma_y = 0, max_iterations = 1000)
} # }