adelie.adelie_core.matrix.MatrixNaiveConvexGatedReluDense32F

class adelie.adelie_core.matrix.MatrixNaiveConvexGatedReluDense32F

Core matrix class for naive convex gated relu matrix with dense underlying.

Methods

__init__(self, mat, mask, n_threads)
bmul(self, arg0, arg1, arg2, arg3, arg4)	Computes a column block matrix-vector multiplication.
bmul_safe(self, arg0, arg1, arg2, arg3, arg4)	Computes a column block matrix-vector multiplication.
btmul(self, arg0, arg1, arg2, arg3)	Computes a column block matrix transpose-vector multiplication increment.
cmul(self, arg0, arg1, arg2)	Computes a column vector-vector multiplication.
cmul_safe(self, arg0, arg1, arg2)	Computes a column vector-vector multiplication.
cols(self)	Returns the number of columns.
cov(self, arg0, arg1, arg2, arg3)	Computes a weighted covariance matrix.
ctmul(self, arg0, arg1, arg2)	Computes a column vector-scalar multiplication increment.
mean(self, arg0, arg1)	Computes the implied column means.
mul(self, arg0, arg1, arg2)	Computes a matrix-vector multiplication.
rows(self)	Returns the number of rows.
sp_tmul(self, arg0, arg1)	Computes a matrix transpose-sparse matrix multiplication.
sq_mul(self, arg0, arg1)	Computes a squared matrix-vector multiplication.
var(self, arg0, arg1, arg2)	Computes the implied column variances.

Attributes

ndim	Number of dimensions.
shape	Shape of the matrix.

__init__(self: adelie.adelie_core.matrix.MatrixNaiveConvexGatedReluDense32F, mat: numpy.ndarray[numpy.float32[m, n], flags.f_contiguous], mask: numpy.ndarray[bool[m, n], flags.f_contiguous], n_threads: int) → None

bmul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: int, arg2: numpy.ndarray[numpy.float32[1, n]], arg3: numpy.ndarray[numpy.float32[1, n]], arg4: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a column block matrix-vector multiplication.

Computes the matrix-vector multiplication (v * w).T @ X[:, j:j+q].

Warning

This function is not thread-safe!

Parameters:

jint

Column index.

qint

Number of columns.

v(n,) ndarray

Vector to multiply with the block matrix.

w(n,) ndarray

Vector of weights.

out(q,) ndarray

Vector to store in-place the result.

bmul_safe(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: int, arg2: numpy.ndarray[numpy.float32[1, n]], arg3: numpy.ndarray[numpy.float32[1, n]], arg4: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a column block matrix-vector multiplication.

Thread-safe version of bmul().

Parameters:

jint

Column index.

qint

Number of columns.

v(n,) ndarray

Vector to multiply with the block matrix.

w(n,) ndarray

Vector of weights.

out(q,) ndarray

Vector to store in-place the result.

btmul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: int, arg2: numpy.ndarray[numpy.float32[1, n]], arg3: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a column block matrix transpose-vector multiplication increment.

Computes the matrix-vector multiplication v.T @ X[:, j:j+q].T. The result is incremented into the output vector.

Warning

This function is not thread-safe!

Parameters:

jint

Column index.

qint

Number of columns.

v(q,) ndarray

Vector to multiply with the block matrix.

out(n,) ndarray

Vector to increment in-place the result.

cmul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: numpy.ndarray[numpy.float32[1, n]], arg2: numpy.ndarray[numpy.float32[1, n]]) → float

Computes a column vector-vector multiplication.

Computes the dot-product (v * w).T @ X[:,j].

Warning

This function is not thread-safe!

Parameters:

jint

Column index.

v(n,) ndarray

Vector to dot product with the j th column.

w(n,) ndarray

Vector of weights.

Returns:

dotfloat

Column vector-vector multiplication.

cmul_safe(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: numpy.ndarray[numpy.float32[1, n]], arg2: numpy.ndarray[numpy.float32[1, n]]) → float

Computes a column vector-vector multiplication.

Thread-safe version of cmul().

Parameters:

jint

Column index.

v(n,) ndarray

Vector to dot product with the j th column.

w(n,) ndarray

Vector of weights.

Returns:

dotfloat

Column vector-vector multiplication.

cols(self: adelie.adelie_core.matrix.MatrixNaiveBase32) → int

Returns the number of columns.

Returns:

colsint

Number of columns.

cov(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: int, arg2: numpy.ndarray[numpy.float32[1, n]], arg3: numpy.ndarray[numpy.float32[m, n], flags.writeable, flags.f_contiguous]) → None

Computes a weighted covariance matrix.

Computes the weighted covariance matrix X[:, j:j+q].T @ W @ X[:, j:j+q].

This function is thread-safe.

Note

Although the name is “covariance”, we do not center the columns of X!

Parameters:

jint

Column index.

qint

Number of columns.

sqrt_weights(n,) ndarray

Square-root of the weights.

out(q, q) ndarray

Matrix to store in-place the result.

ctmul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: int, arg1: float, arg2: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a column vector-scalar multiplication increment.

Computes the vector-scalar multiplication v * X[:,j]. The result is incremented into the output vector.

Warning

This function is not thread-safe!

Parameters:

jint

Column index.

vfloat

Scalar to multiply with the j th column.

out(n,) ndarray

Vector to increment in-place the result.

mean(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: numpy.ndarray[numpy.float32[1, n]], arg1: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes the implied column means.

The default implied column means are given by w.T @ X. Unless stated otherwise, this function will compute the default version.

Parameters:

weights(n,) ndarray

Vector of weights.

out(p,) ndarray

Vector to store in-place the result.

mul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: numpy.ndarray[numpy.float32[1, n]], arg1: numpy.ndarray[numpy.float32[1, n]], arg2: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a matrix-vector multiplication.

Computes the matrix-vector multiplication (v * w).T @ X.

Parameters:

v(n,) ndarray

Vector to multiply with the matrix.

w(n,) ndarray

Vector of weights.

out(p,) ndarray

Vector to store in-place the result.

rows(self: adelie.adelie_core.matrix.MatrixNaiveBase32) → int

Returns the number of rows.

Returns:

rowsint

Number of rows.

sp_tmul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: scipy.sparse.csr_matrix[numpy.float32], arg1: numpy.ndarray[numpy.float32[m, n], flags.writeable, flags.c_contiguous]) → None

Computes a matrix transpose-sparse matrix multiplication.

Computes the matrix transpose-sparse matrix multiplication v.T @ X.T.

Parameters:

v(L, p) csr_matrix

Sparse matrix to multiply with the matrix.

out(L, n) ndarray

Matrix to store in-place the result.

sq_mul(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: numpy.ndarray[numpy.float32[1, n]], arg1: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes a squared matrix-vector multiplication.

Computes the squared matrix-vector multiplication w.T @ X ** 2.

Parameters:

weights(n,) ndarray

Vector of weights.

out(p,) ndarray

Vector to store in-place the result.

var(self: adelie.adelie_core.matrix.MatrixNaiveBase32, arg0: numpy.ndarray[numpy.float32[1, n]], arg1: numpy.ndarray[numpy.float32[1, n]], arg2: numpy.ndarray[numpy.float32[1, n], flags.writeable]) → None

Computes the implied column variances.

The default implied column variances are given by w.T @ (X-c[None])**2. Unless stated otherwise, this function will compute the default version.

Parameters:

centers(p,) ndarray

Vector of centers.

weights(n,) ndarray

Vector of weights.

out(p,) ndarray

Vector to store in-place the result.

ndim

Number of dimensions. It is always 2.

shape

Shape of the matrix.