mlpack

Core math utilities

mlpack provides a number of mathematical utility classes and functions on top of Armadillo.

🔗 Aliases

Aliases are matrix, vector, or cube objects that share memory with another matrix, vector, or cube. They are often used internally inside of mlpack to avoid copies.

Important caveats about aliases:

🔗 Range

The Range class represents a simple mathematical range (i.e. [0, 3]), with the bounds represented as doubles.

🔗 Constructors

🔗 Accessing and modifying range properties

🔗 Working with ranges

🔗 Usage example

mlpack::Range r1(5.0, 6.0); // [5, 6]
mlpack::Range r2(7.0, 8.0); // [7, 8]

mlpack::Range r3 = r1 | r2; // [5, 8]
mlpack::Range r4 = r1 & r2; // empty range

bool b1 = r1.Contains(r2); // false
bool b2 = r1.Contains(5.5); // true
bool b3 = r1.Contains(r3); // true
bool b4 = r3.Contains(r4); // false

// Create a range of `float`s and a range of `int`s.
mlpack::RangeType<float> r5(1.0f, 1.5f); // [1.0, 1.5]
mlpack::RangeType<int> r6(3, 4); // [3, 4]

Range is used by:

🔗 ColumnCovariance()

Example:

// Generate a random data matrix with 100 points in 5 dimensions.
arma::mat data(5, 100, arma::fill::randu);

// Compute the covariance matrix of the column-major matrix.
arma::mat cov = mlpack::ColumnCovariance(data);
cov.print("Covariance of random matrix:");

🔗 ColumnsToBlocks

The ColumnsToBlocks class provides a way to transform data points (e.g. columns in a matrix) into a block matrix format, primarily useful for visualization as an image.

As a simple example, given a matrix with four columns A, B, C, and D, ColumnsToBlocks can transform this matrix into the form

[[m m m m m]
 [m A m B m]
 [m m m m m]
 [m C m D m]
 [m m m m m]]

where m is a margin, and where each column may itself be reshaped into a block.

🔗 Constructors

🔗 Properties

🔗 Scaling values

ColumnsToBlocks also has the capability of linearly scaling values of the inputs to a given range.

Note: the margin element (ctb.BufValue()) is considered during the scaling process.

🔗 Transforming into block format

🔗 Examples

Reshape two 4-element vectors into one row of two blocks.

// This matrix has two columns.
arma::mat input;
input = { { -1.0000, 0.1429 },
          { -0.7143, 0.4286 },
          { -0.4286, 0.7143 },
          { -0.1429, 1.0000 } };
input.print("Input columns:");

arma::mat output;
mlpack::ColumnsToBlocks ctb(1, 2);
ctb.Transform(input, output);

// The columns of the input will be reshaped as a square which is
// surrounded by padding value -1 (this value could be changed with the
// BufValue() method):
// -1.0000  -1.0000  -1.0000  -1.0000  -1.0000  -1.0000  -1.0000
// -1.0000  -1.0000  -0.4286  -1.0000   0.1429   0.7143  -1.0000
// -1.0000  -0.7143  -0.1429  -1.0000   0.4286   1.0000  -1.0000
// -1.0000  -1.0000  -1.0000  -1.0000  -1.0000  -1.0000  -1.0000
output.print("Output using 2x2 block size:");

// Now, let's change some parameters; let's have each input column output not
// as a square, but as a 4x1 vector.
ctb.BlockWidth(1);
ctb.BlockHeight(4);
ctb.Transform(input, output);

// The output here will be similar, but each maximal input is 4x1:
// -1.0000 -1.0000 -1.0000 -1.0000 -1.0000
// -1.0000 -1.0000 -1.0000  0.1429 -1.0000
// -1.0000 -0.7143 -1.0000  0.4286 -1.0000
// -1.0000 -0.4286 -1.0000  0.7143 -1.0000
// -1.0000 -0.1429 -1.0000  1.0000 -1.0000
// -1.0000 -1.0000 -1.0000 -1.0000 -1.0000
output.print("Output using 4x1 block size:");

Load simple images and reshape into blocks.

// Load some favicons from websites associated with mlpack.
std::vector<std::string> images;
// See the following files:
// - https://datasets.mlpack.org/images/mlpack-favicon.png
// - https://datasets.mlpack.org/images/ensmallen-favicon.png
// - https://datasets.mlpack.org/images/armadillo-favicon.png 
// - https://datasets.mlpack.org/images/bandicoot-favicon.png
images.push_back("mlpack-favicon.png");
images.push_back("ensmallen-favicon.png");
images.push_back("armadillo-favicon.png");
images.push_back("bandicoot-favicon.png");

mlpack::data::ImageInfo info;
info.Channels() = 1; // Force loading in grayscale.

arma::mat matrix;
mlpack::data::Load(images, matrix, info, true);

// Now `matrix` has 4 columns, each of which is an individual image.
// Let's save that as its own image just for visualization.
mlpack::data::ImageInfo outInfo(matrix.n_cols, matrix.n_rows, 1);
mlpack::data::Save("favicons-matrix.png", matrix, outInfo, true);

// Use ColumnsToBlocks to create a 2x2 block matrix holding each image.
mlpack::ColumnsToBlocks ctb(2, 2);
ctb.BufValue(0.0); // Use 0 for the margin value.
ctb.BufSize(2); // Use 2-pixel margins.

arma::mat blocks;
ctb.Transform(matrix, blocks);

mlpack::data::ImageInfo blockOutInfo(blocks.n_cols, blocks.n_rows, 1);
mlpack::data::Save("favicons-blocks.png", blocks, blockOutInfo, true);

The resulting images (before and after using ColumnsToBlocks) are shown below.

Before:

four favicons each as a column in a matrix, unintelligible

After:

four favicons each as a block in a larger image, much better

🔗 See also

🔗 Distribution utilities

Example:

const double d1 = mlpack::Digamma(0.25);
const double d2 = mlpack::Digamma(1.0);

const double t1 = mlpack::Trigamma(0.25);
const double t2 = mlpack::Trigamma(1.0);

std::cout << "Digamma(0.25):  " << d1 << "." << std::endl;
std::cout << "Digamma(1.0):   " << d2 << "." << std::endl;
std::cout << "Trigamma(0.25): " << t1 << "." << std::endl;
std::cout << "Trigamma(1.0):  " << t2 << "." << std::endl;

🔗 RandVector()

Example:

// Generate a random 10-dimensional vector.
arma::vec v;
v.set_size(10);
RandVector(v);
v.print("Random 10-dimensional vector: ");

std::cout << "Random 10-dimensional vector: " << std::endl;
std::cout << v.t();
std::cout << "L2-norm of vector (should be 1): " << arma::norm(v, 2) << "."
    << std::endl;

🔗 Logarithmic utilities

mlpack contains a few functions that are useful for working with logarithms, or vectors containing logarithms.

🔗 MultiplyCube2Cube()

Example usage:

// Generate two random cubes.
arma::cube x(10, 100, 5, arma::fill::randu); // 5 matrices, each 10x100.
arma::cube y(12, 100, 5, arma::fill::randu); // 5 matrices, each 12x100.

arma::cube z = mlpack::MultiplyCube2Cube(x, y, false, true);

// Output size should be 10x12x5.
std::cout << "Output size: " << z.n_rows << "x" << z.n_cols << "x" << z.n_slices
    << "." << std::endl;

🔗 MultiplyMat2Cube()

Example usage:

// Generate random inputs.
arma::mat  x(10, 100,    arma::fill::randu); // Random 10x100 matrix.
arma::cube y(12, 100, 5, arma::fill::randu); // 5 matrices, each 12x100.

arma::cube z = mlpack::MultiplyMat2Cube(x, y, false, true);

// Output size should be 10x12x5.
std::cout << "Output size: " << z.n_rows << "x" << z.n_cols << "x" << z.n_slices
    << "." << std::endl;

🔗 MultiplyCube2Mat()

Example usage:

// Generate two random cubes.
arma::cube x(12, 50, 5, arma::fill::randu); // 5 matrices, each 12x50.
arma::mat  y(12, 60,    arma::fill::randu); // Random 12x60 matrix.

arma::cube z = mlpack::MultiplyCube2Mat(x, y, true, false);

// Output size should be 50x60x5.
std::cout << "Output size: " << z.n_rows << "x" << z.n_cols << "x" << z.n_slices
    << "." << std::endl;

🔗 Quantile()

Example usage:

// 70% of points from N(0, 1) are less than q1 = 0.524.
double q1 = mlpack::Quantile(0.7);

// 90% of points from N(0, 1) are less than q2 = 1.282.
double q2 = mlpack::Quantile(0.9);

// 50% of points from N(1, 1) are less than q3 = 1.0.
double q3 = mlpack::Quantile(0.5, 1.0); // Quantile of 1.0 for N(1, 1) is 1.0.

// 10% of points from N(1, 0.1) are less than q4 = 0.871.
double q4 = mlpack::Quantile(0.1, 1.0, 0.1);

std::cout << "Quantile(0.7): " << q1 << "." << std::endl;
std::cout << "Quantile(0.9): " << q2 << "." << std::endl;
std::cout << "Quantile(0.5, 1.0): " << q3 << "." << std::endl;
std::cout << "Quantile(0.1, 1.0, 0.1): " << q4 << "." << std::endl;

🔗 RNG and random number utilities

On top of the random number generation support that Armadillo provides via randu(), randn(), and randi(), mlpack provides a few additional thread-safe random number generation functions for generating random scalar values.

Examples:

mlpack::RandomSeed(123); // Set a specific random seed.

const double r1 = mlpack::Random();             // In the range [0, 1).
const double r2 = mlpack::Random(3, 4);         // In the range [3, 4).
const double r3 = mlpack::RandBernoulli(0.25);  // P(1) = 0.25.
const int    r4 = mlpack::RandInt(10);          // In the range [0, 10).
const int    r5 = mlpack::RandInt(5, 10);       // In the range [5, 10).
const double r6 = mlpack::RandNormal();         // r6 ~ N(0, 1).
const double r7 = mlpack::RandNormal(2.0, 3.0); // r7 ~ N(2, 3).

std::cout << "Random():            " << r1 << "." << std::endl;
std::cout << "Random(3, 4):        " << r2 << "." << std::endl;
std::cout << "RandBernoulli(0.25): " << r3 << "." << std::endl;
std::cout << "RandInt(10):         " << r4 << "." << std::endl;
std::cout << "RandInt(5, 10):      " << r5 << "." << std::endl;
std::cout << "RandNormal():        " << r6 << "." << std::endl;
std::cout << "RandNormal(2, 3):    " << r7 << "." << std::endl;

🔗 RandomBasis()

The RandomBasis() function generates a random d-dimensional orthogonal basis.

Example:

arma::mat basis;

// Generate a 10-dimensional random basis.
mlpack::RandomBasis(basis, 10);

// Each two vectors are orthogonal.
std::cout << "Dot product of basis vectors 2 and 4: "
    << arma::dot(basis.col(2), basis.col(4))
    << " (should be zero or very close!)." << std::endl;

🔗 ShuffleData()

Shuffle a column-major dataset and associated labels/responses, optionally with weights. This preserves the connection of each data point to its label (and optionally its weight).

Note: when inputData is a cube (e.g. arma::cube or similar), the columns of the cube will be shuffled.

Example usage:

// See https://datasets.mlpack.org/iris.csv.
arma::mat dataset;
mlpack::data::Load("iris.csv", dataset, true);
// See https://datasets.mlpack.org/iris.labels.csv.
arma::Row<size_t> labels;
mlpack::data::Load("iris.labels.csv", labels, true);

// Now shuffle the points in the iris dataset.
arma::mat shuffledDataset;
arma::Row<size_t> shuffledLabels;
mlpack::ShuffleData(dataset, labels, shuffledDataset, shuffledLabels);

std::cout << "Before shuffling, the first point was: " << std::endl;
std::cout << "  " << dataset.col(0).t();
std::cout << "with label " << labels[0] << "." << std::endl;
std::cout << std::endl;
std::cout << "After shuffling, the first point is: " << std::endl;
std::cout << "  " << shuffledDataset.col(0).t();
std::cout << "with label " << shuffledLabels[0] << "." << std::endl;

// Generate random weights, then shuffle those also.
arma::rowvec weights(dataset.n_cols, arma::fill::randu);
arma::rowvec shuffledWeights;
mlpack::ShuffleData(dataset, labels, weights, shuffledDataset, shuffledLabels,
    shuffledWeights);

std::cout << std::endl << std::endl;
std::cout << "Before shuffling with weights, the first point was: "
    << std::endl;
std::cout << "  " << dataset.col(0).t();
std::cout << "with label " << labels[0] << " and weight " << weights[0] << "."
    << std::endl;
std::cout << std::endl;
std::cout << "After shuffling with weights, the first point is: " << std::endl;
std::cout << "  " << shuffledDataset.col(0).t();
std::cout << "with label " << shuffledLabels[0] << " and weight "
    << shuffledWeights[0] << "." << std::endl;