Struct xgboost::DMatrix

pub struct DMatrix { /* fields omitted */ }

Data matrix used throughout XGBoost for training/predicting Booster models.

It's used as a container for both features (i.e. a row for every instance), and an optional true label for that instance (as an f32 value).

Can be created files, or from dense or sparse (CSR or CSC) matrices.

Examples

Load from file

Load matrix from file in LIBSVM or binary format.

use xgboost::DMatrix;

let dmat = DMatrix::load("somefile.txt").unwrap();

Create from dense array

use xgboost::DMatrix;

let data = &[1.0, 0.5, 0.2, 0.2,
             0.7, 1.0, 0.1, 0.1,
             0.2, 0.0, 0.0, 1.0];
let num_rows = 3;
let mut dmat = DMatrix::from_dense(data, num_rows).unwrap();
assert_eq!(dmat.shape(), (3, 4));

// set true labels for each row
dmat.set_labels(&[1.0, 0.0, 1.0]);

Create from sparse CSR matrix

Create from sparse representation of

[[1.0, 0.0, 2.0],
 [0.0, 0.0, 3.0],
 [4.0, 5.0, 6.0]]

use xgboost::DMatrix;

let indptr = &[0, 2, 3, 6];
let indices = &[0, 2, 2, 0, 1, 2];
let data = &[1.0, 2.0, 3.0, 4.0, 5.0, 6.0];
let dmat = DMatrix::from_csr(indptr, indices, data, None).unwrap();
assert_eq!(dmat.shape(), (3, 3));

Methods

impl DMatrix

pub fn from_dense(data: &[f32], num_rows: usize) -> XGBResult<Self>

Create a new DMatrix from dense array in row-major order.

E.g. the matrix

[[1.0, 2.0],
 [3.0, 4.0],
 [5.0, 6.0]]

would be represented converted into a DMatrix with

use xgboost::DMatrix;

let data = &[1.0, 2.0, 3.0, 4.0, 5.0, 6.0];
let num_rows = 3;
let dmat = DMatrix::from_dense(data, num_rows).unwrap();

pub fn from_csr(
    indptr: &[usize],
    indices: &[usize],
    data: &[f32],
    num_cols: Option<usize>
) -> XGBResult<Self>

Create a new DMatrix from a sparse CSR matrix.

Uses standard CSR representation where the column indices for row i are stored in indices[indptr[i]:indptr[i+1]] and their corresponding values are stored in data[indptr[i]:indptr[i+1].

If num_cols is set to None, number of columns will be inferred from given data.

pub fn from_csc(
    indptr: &[usize],
    indices: &[usize],
    data: &[f32],
    num_rows: Option<usize>
) -> XGBResult<Self>

Create a new DMatrix from a sparse CSC) matrix.

Uses standard CSC representation where the row indices for column i are stored in indices[indptr[i]:indptr[i+1]] and their corresponding values are stored in data[indptr[i]:indptr[i+1].

If num_rows is set to None, number of rows will be inferred from given data.

pub fn load<P: AsRef<Path>>(path: P) -> XGBResult<Self>

Create a new DMatrix from given file.

Supports text files in LIBSVM format, CSV, binary files written either by save, or from another XGBoost library.

For more details on accepted formats, seem the XGBoost input format documentation.

LIBSVM format

Specified data in a sparse format as:

<label> <index>:<value> [<index>:<value> ...]

E.g.

0 1:1 9:0 11:0
1 9:1 11:0.375 15:1
0 1:0 8:0.22 11:1

pub fn save<P: AsRef<Path>>(&self, path: P) -> XGBResult<()>

Serialise this DMatrix as a binary file to given path.

pub fn num_rows(&self) -> usize

Get the number of rows in this matrix.

pub fn num_cols(&self) -> usize

Get the number of columns in this matrix.

pub fn shape(&self) -> (usize, usize)

Get the shape (rows x columns) of this matrix.

pub fn slice(&self, indices: &[usize]) -> XGBResult<DMatrix>

Get a new DMatrix as a containing only given indices.

pub fn get_root_index(&self) -> XGBResult<&[u32]>

Gets the specified root index of each instance, can be used for multi task setting.

See the XGBoost documentation for more information.

pub fn set_root_index(&mut self, array: &[u32]) -> XGBResult<()>

Sets the specified root index of each instance, can be used for multi task setting.

See the XGBoost documentation for more information.

pub fn get_labels(&self) -> XGBResult<&[f32]>

Get ground truth labels for each row of this matrix.

pub fn set_labels(&mut self, array: &[f32]) -> XGBResult<()>

Set ground truth labels for each row of this matrix.

pub fn get_weights(&self) -> XGBResult<&[f32]>

Get weights of each instance.

pub fn set_weights(&mut self, array: &[f32]) -> XGBResult<()>

Set weights of each instance.

pub fn get_base_margin(&self) -> XGBResult<&[f32]>

Get base margin.

pub fn set_base_margin(&mut self, array: &[f32]) -> XGBResult<()>

Set base margin.

If specified, xgboost will start from this margin, can be used to specify initial prediction to boost from.

pub fn set_group(&mut self, group: &[u32]) -> XGBResult<()>

Set the index for the beginning and end of a group.

Needed when the learning task is ranking.

See the XGBoost documentation for more information.

Trait Implementations

impl Drop for DMatrix

fn drop(&mut self)

Executes the destructor for this type.