Package gmisclib :: Module matrix_rearrange_labels

Module matrix_rearrange_labels

This module helps you plot confusion matrices or similar images where the axis labels do not have a natural order. So, the probability of confusing two phonemes is a perfect example: phonemes do not naturally fall onto a 1-dimensional sequence, so one is free to put them in any order one likes. Given that, one might as well put them into an order that reveals something interesting about the probabilities.

All the functions ending in "2" work on rectangular arrays. All the functions without "2" work only on square arrays, and they assume that that both axes remain in the same order.

Classes
	diagfom A class that defines a figure-of-merit for a matrix.
	diagfom2 This tries to make your matrix approximately diagonal for rectangular matrices.
	blockfom This tries to make your matrix into a block form.

Functions

symm_swap_toward_minimal_fom(m, fom, maxtries)
Used internally.

source code

swap_toward_minimal_fom(m, fom, maxtries)
Used internally.

source code

tuple(numpy.ndarray, list(something))

swap_toward_diag(m, lbls, maxtries=None, sign=1)
Swap the rows and columns of a matrix to bring it closer to a diagonal matrix: i.e.

source code

tuple(numpy.ndarray, list(something))

swap_toward_positive(m, lbls, maxtries=None, sign=1)
Swap the rows and columns of a matrix to put the most positive values on the main diagonal.

source code

tuple(numpy.ndarray, list(something), list(something))

swap_toward_diag2(m, lbl1, lbl2, maxtries=None)
Swap the rows and columns of a matrix to make it roughly diagonal: i.e.

source code

pos_near_diag2(m, lbl1, lbl2, maxtries=None)

source code

neg_near_diag2(m, lbl1, lbl2, maxtries=None)

source code

tuple(numpy.ndarray, list(something), list(something))

swap_toward_blocks2(m, lbl1, lbl2, maxtries=None)
Swap rows and columns of a matrix to bring it closer to a block form, where similar values occur together in blocks.

source code

tuple(numpy.ndarray), list(something))

swap_toward_blocks(m, lbls, maxtries=None)
Swap rows and columns of a matrix to bring it closer to a block form, where similar values occur together in blocks.

source code

test1()
Two test cases on 2x2 matrices.

source code

test()

source code

Variables
	__package__ = `'gmisclib'`

Imports: numpy, die, blue_data_selector

Function Details

symm_swap_toward_minimal_fom(m, fom, maxtries)

source code

Used internally. Finds an ordering of the labels that minimizes whatever fom you supply. Both axes will have the same order.

swap_toward_minimal_fom(m, fom, maxtries)

source code

Used internally. Finds an ordering of the labels that minimizes whatever fom you supply. The ordering of the two axes may be different.

swap_toward_diag(m, lbls, maxtries=None, sign=1)

source code

Swap the rows and columns of a matrix to bring it closer to a diagonal matrix: i.e. entries with large absolute values on the main diagonal and small entries away from the main diagonal. Rows and columns are swapped together, so that the ordering or rows will match the ordering of columns.

Parameters:

m (numpy.ndarray) - a 2-dimensional array
lbls (list(anything)) - a list of arbitrary labels for each row or column.
maxtries (int or None) - how hard to work at optimizing the matrix layout? None gives a resonable default value. Appropriate values are a few times the number of elements in the matrix.
sign (int) - (default=1) if sign=-1, do the opposite: put the small absolute values on the main diagonal.

Returns: tuple(numpy.ndarray, list(something))

A tuple containing the swapped matrix and a swapped list of values.

swap_toward_positive(m, lbls, maxtries=None, sign=1)

source code

Swap the rows and columns of a matrix to put the most positive values on the main diagonal. Rows and columns are swapped together, so that the ordering or rows will match the ordering of columns.

Parameters:

m (numpy.ndarray) - a 2-dimensional array
lbls (list(anything)) - a list of arbitrary labels for each row or column.
maxtries (int or None) - how hard to work at optimizing the matrix layout? None gives a resonable default value. Appropriate values are a few times the number of elements in the matrix.
sign (int) - (default=1) if sign=-1, do the opposite: put the negative entries on the main diagonal.

Returns: tuple(numpy.ndarray, list(something))

A tuple containing the swapped matrix and a swapped list of labels.

swap_toward_diag2(m, lbl1, lbl2, maxtries=None)

source code

Swap the rows and columns of a matrix to make it roughly diagonal: i.e. large entries on the main diagonal and small entries away from the main diagonal. Note that this will work even if the matrix is not square.

Parameters:

m (numpy.ndarray) - a 2-dimensional array
lbl1 (list(anything)) - a list of arbitrary labels for the first index of the matrix m.
lbl2 (list(anything)) - a list of arbitrary labels for the second index of the matrix m.
maxtries (int or None) - how hard to work at optimizing the matrix layout? None gives a resonable default value. Appropriate values are a few times the number of elements in the matrix.

Returns: tuple(numpy.ndarray, list(something), list(something))

A tuple containing the swapped matrix and the two swapped lists of labels, one for the first and one for the second axis.

swap_toward_blocks2(m, lbl1, lbl2, maxtries=None)

source code

Swap rows and columns of a matrix to bring it closer to a block form, where similar values occur together in blocks.

Parameters:

m (numpy.ndarray) - a 2-dimensional array
lbl1 (list(anything)) - a list of arbitrary labels for the first axis of m
lbl2 (list(anything)) - a list of arbitrary labels for the second axis of m
maxtries (int or None) - how hard to work at optimizing the matrix layout? None gives a resonable default value. Appropriate values are a few times the number of elements in the matrix.

Returns: tuple(numpy.ndarray, list(something), list(something))

A tuple containing the swapped matrix and the two swapped lists of values, one for the first and one for the second axis.

swap_toward_blocks(m, lbls, maxtries=None)

source code

Swap rows and columns of a matrix to bring it closer to a block form, where similar values occur together in blocks. It minimizes the sum of changes when proceeding along a row or column.

Parameters:

m (numpy.ndarray) - a 2-dimensional array
lbls (list(anything)) - a list of arbitrary labels for the first axis of m
maxtries (int or None) - how hard to work at optimizing the matrix layout? None gives a resonable default value. Appropriate values are a few times the number of elements in the matrix.

Returns: tuple(numpy.ndarray), list(something))

A tuple containing the swapped matrix and a swapped lists of values.