nerva_numpy.matrix_operations

Matrix operations built on top of torch to support the math in the library.

The functions here intentionally mirror the names in the accompanying docs. They operate on 1D/2D tensors and keep broadcasting explicit for clarity.

Functions

Diag(x)	Create diagonal matrix with x as diagonal.
abs(X)	Element-wise absolute value |X|.
apply(f, X)	Element-wise application of function f to X.
clip(X, a_min, a_max)	Element-wise clip of X to [a_min, a_max].
column_repeat(x, n)	Repeat column vector x horizontally n times.
columns_max(X)	Returns a column vector with the maximum values of each row in X.
columns_mean(X)	Returns a column vector with the mean values of each row in X.
columns_sum(X)	Sum over columns (returns row vector).
diag(X)	Extract diagonal of X as a vector.
dot(x, y)	Dot product of vectors x and y.
elements_sum(X)	Returns the sum of the elements of X.
exp(X)	Element-wise exponential exp(X).
hadamard(X, Y)	Element-wise product X ⊙ Y.
identity(n)	Returns the nxn identity matrix.
inv_sqrt(X)	Element-wise inverse square root X^(-1/2) with epsilon for stability.
is_column_vector(x)	Check if x can be treated as a column vector.
is_row_vector(x)	Check if x can be treated as a row vector.
is_square(X)	Check if X is a square matrix.
is_vector(x)	Check if x is a 1D tensor.
log(X)	Element-wise natural logarithm log(X).
log_sigmoid(X)	Element-wise log(sigmoid(X)) computed stably.
ones(m[, n])	Returns an mxn matrix with all elements equal to 1.
product(X, Y)	Matrix multiplication X @ Y.
reciprocal(X)	Element-wise reciprocal 1/X.
row_repeat(x, m)	Repeat row vector x vertically m times.
rows_max(X)	Returns a row vector with the maximum values of each column in X.
rows_mean(X)	Returns a row vector with the mean values of each column in X.
rows_sum(X)	Sum over rows (returns column vector).
sqrt(X)	Element-wise square root √X.
square(X)	Element-wise square X².
vector_size(x)	Get size along first dimension.
zeros(m[, n])	Returns an mxn matrix with all elements equal to 0.

nerva_numpy.matrix_operations.is_vector(x: numpy.ndarray) → bool

Check if x is a 1D tensor.

nerva_numpy.matrix_operations.is_column_vector(x: numpy.ndarray) → bool

Check if x can be treated as a column vector.

nerva_numpy.matrix_operations.is_row_vector(x: numpy.ndarray) → bool

Check if x can be treated as a row vector.

nerva_numpy.matrix_operations.vector_size(x: numpy.ndarray) → int

Get size along first dimension.

nerva_numpy.matrix_operations.is_square(X: numpy.ndarray) → bool

Check if X is a square matrix.

nerva_numpy.matrix_operations.dot(x: numpy.ndarray, y: numpy.ndarray)

Dot product of vectors x and y.

nerva_numpy.matrix_operations.zeros(m: int, n=None) → numpy.ndarray

Returns an mxn matrix with all elements equal to 0.

nerva_numpy.matrix_operations.ones(m: int, n=None) → numpy.ndarray

Returns an mxn matrix with all elements equal to 1.

nerva_numpy.matrix_operations.identity(n: int) → numpy.ndarray

Returns the nxn identity matrix.

nerva_numpy.matrix_operations.product(X: numpy.ndarray, Y: numpy.ndarray) → numpy.ndarray

Matrix multiplication X @ Y.

nerva_numpy.matrix_operations.hadamard(X: numpy.ndarray, Y: numpy.ndarray) → numpy.ndarray

Element-wise product X ⊙ Y.

nerva_numpy.matrix_operations.diag(X: numpy.ndarray) → numpy.ndarray

Extract diagonal of X as a vector.

nerva_numpy.matrix_operations.Diag(x: numpy.ndarray) → numpy.ndarray

Create diagonal matrix with x as diagonal.

nerva_numpy.matrix_operations.elements_sum(X: numpy.ndarray)

Returns the sum of the elements of X.

nerva_numpy.matrix_operations.column_repeat(x: numpy.ndarray, n: int) → numpy.ndarray

Repeat column vector x horizontally n times.

nerva_numpy.matrix_operations.row_repeat(x: numpy.ndarray, m: int) → numpy.ndarray

Repeat row vector x vertically m times.

nerva_numpy.matrix_operations.columns_sum(X: numpy.ndarray) → numpy.ndarray

Sum over columns (returns row vector).

nerva_numpy.matrix_operations.rows_sum(X: numpy.ndarray) → numpy.ndarray

Sum over rows (returns column vector).

nerva_numpy.matrix_operations.columns_max(X: numpy.ndarray) → numpy.ndarray

Returns a column vector with the maximum values of each row in X.

nerva_numpy.matrix_operations.rows_max(X: numpy.ndarray) → numpy.ndarray

Returns a row vector with the maximum values of each column in X.

nerva_numpy.matrix_operations.columns_mean(X: numpy.ndarray) → numpy.ndarray

Returns a column vector with the mean values of each row in X.

nerva_numpy.matrix_operations.rows_mean(X: numpy.ndarray) → numpy.ndarray

Returns a row vector with the mean values of each column in X.

nerva_numpy.matrix_operations.apply(f, X: numpy.ndarray) → numpy.ndarray

Element-wise application of function f to X.

nerva_numpy.matrix_operations.exp(X: numpy.ndarray) → numpy.ndarray

Element-wise exponential exp(X).

nerva_numpy.matrix_operations.log(X: numpy.ndarray) → numpy.ndarray

Element-wise natural logarithm log(X).

nerva_numpy.matrix_operations.reciprocal(X: numpy.ndarray) → numpy.ndarray

Element-wise reciprocal 1/X.

nerva_numpy.matrix_operations.square(X: numpy.ndarray) → numpy.ndarray

Element-wise square X².

nerva_numpy.matrix_operations.sqrt(X: numpy.ndarray) → numpy.ndarray

Element-wise square root √X.

nerva_numpy.matrix_operations.inv_sqrt(X: numpy.ndarray) → numpy.ndarray

Element-wise inverse square root X^(-1/2) with epsilon for stability.

nerva_numpy.matrix_operations.log_sigmoid(X: numpy.ndarray) → numpy.ndarray

Element-wise log(sigmoid(X)) computed stably.

nerva_numpy.matrix_operations.abs(X: numpy.ndarray) → numpy.ndarray

Element-wise absolute value |X|.

nerva_numpy.matrix_operations.clip(X: numpy.ndarray, a_min: float, a_max: float) → numpy.ndarray

Element-wise clip of X to [a_min, a_max].