skcriteria.cmp.ranks_cmp module

Ranking comparison routines.

class skcriteria.cmp.ranks_cmp.RanksComparator(ranks)

Bases: Sequence, DiffEqualityMixin

Rankings comparator object.

This class is intended to contain a collection of rankings on which you want to do comparative analysis.

All rankings must have exactly the same alternatives, although their order may vary.

All methods support the untied parameter, which serves to untie rankings in case there are results that can assign more than one alternative to the same position (e.g.``ELECTRE2``).

Parameters:

ranks (list) – List of (name, ranking) tuples of skcriteria.agg.RankResult with the same alternatives.

See also

skcriteria.cmp.mkrank_cmp

Convenience function for simplified ranks comparator construction.

property ranks

List of ranks in the comparator.

property named_ranks

Dictionary-like object, with the following attributes.

Read-only attribute to access any rank parameter by user given name. Keys are ranks names and values are rannks parameters.

diff(other, rtol=1e-05, atol=1e-08, equal_nan=True, check_dtypes=False)

Return the difference between two objects within a tolerance.

This method should be implemented by subclasses to define how differences between objects are calculated.

The tolerance parameters rtol and atol, equal_nan, and check_dtypes are provided to be used by the numpy and pandas equality functions. These parameters allow you to customize the behavior of the equality comparison, such as setting the relative and absolute tolerance for numeric comparisons, considering NaN values as equal, and checking for the data type of the objects being compared.

Parameters:

• other (object) – The object to compare to.

• rtol (float, optional) – The relative tolerance parameter. Default is 1e-05.

• atol (float, optional) – The absolute tolerance parameter. Default is 1e-08.

• equal_nan (bool, optional) – Whether to consider NaN values as equal. Default is True.

• check_dtypes (bool, optional) – Whether to check the data type of the objects. Default is False.

Returns:

The difference between the current and the other object.

Return type:

the_diff

See also

equals, aequals, numpy.isclose(), numpy.all(), numpy.any(), numpy.equal(), numpy.allclose()

Notes

The tolerance values are positive, typically very small numbers. The relative difference (rtol * abs(b)) and the absolute difference atol are added together to compare against the absolute difference between a and b.

NaNs are treated as equal if they are in the same place and if equal_nan=True. Infs are treated as equal if they are in the same place and of the same sign in both arrays.

to_dataframe(*, untied=False)

Convert the entire RanksComparator into a dataframe.

The alternatives are the rows, and the different rankings are the columns.

Parameters:

untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

Returns:

A RanksComparator as pandas DataFrame.

Return type:

pd.DataFrame

corr(*, untied=False, **kwargs)

Compute pairwise correlation of rankings, excluding NA/null values.

By default the pearson correlation coefficient is used.

Please check the full documentation of a pandas.DataFrame.corr() method for details about the implementation.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the pandas.DataFrame.corr() method.

Returns:

A DataFrame with the correlation between rankings.

Return type:

pd.DataFrame

cov(*, untied=False, **kwargs)

Compute pairwise covariance of rankings, excluding NA/null values.

Please check the full documentation of a pandas.DataFrame.cov() method for details about the implementation.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the pandas.DataFrame.cov() method.

Returns:

A DataFrame with the covariance between rankings.

Return type:

pd.DataFrame

r2_score(*, untied=False, **kwargs)

Compute pairwise coefficient of determination regression score function of rankings, excluding NA/null values.

Best possible score is 1.0 and it can be negative (because the model can be arbitrarily worse).

Please check the full documentation of a sklearn.metrics.r2_score function for details about the implementation and the behaviour.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the sklearn.metrics.r2_score() function.

Returns:

A DataFrame with the coefficient of determination between rankings.

Return type:

pd.DataFrame

distance(*, untied=False, metric='hamming', **kwargs)

Compute pairwise distance between rankings.

By default the ‘hamming’ distance is used, which is simply the proportion of disagreeing components in Two rankings.

Please check the full documentation of a scipy.spatial.distance.pdist function for details about the implementation and the behaviour.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• metric (str or function, default "hamming") – The distance metric to use. The distance function can be ‘braycurtis’, ‘canberra’, ‘chebyshev’, ‘cityblock’, ‘correlation’, ‘cosine’, ‘dice’, ‘euclidean’, ‘hamming’, ‘jaccard’, ‘jensenshannon’, ‘kulczynski1’, ‘mahalanobis’, ‘matching’, ‘minkowski’, ‘rogerstanimoto’, ‘russellrao’, ‘seuclidean’, ‘sokalmichener’, ‘sokalsneath’, ‘sqeuclidean’, ‘yule’.

• kwargs – Other keyword arguments are passed to the scipy.spatial.distance.pdist() function.

Returns:

A DataFrame with the distance between rankings.

Return type:

pd.DataFrame

property plot

Plot accessor.

class skcriteria.cmp.ranks_cmp.RanksComparatorPlotter(ranks_cmp)

Bases: AccessorABC

RanksComparator plot utilities.

Kind of plot to produce:

• ‘flow’ : Changes in the rankings of the alternatives as flow lines (default)

• ‘reg’ : Pairwise rankings data and a linear regression model fit plot.

• ‘heatmap’ : Rankings as a color-encoded matrix.

• ‘corr’ : Pairwise correlation of rankings as a color-encoded matrix.

• ‘cov’ : Pairwise covariance of rankings as a color-encoded matrix.

• ‘r2_score’ : Pairwise coefficient of determination regression score function of rankings as a color-encoded matrix.

• ‘distance’ : Pairwise distance between rankings as a color-encoded matrix.

• ‘box’ : Box-plot of rankings with respect to alternatives

• ‘bar’ : Ranking of alternatives by method with vertical bars.

• ‘barh’ : Ranking of alternatives by method with horizontal bars.

flow(*, untied=False, grid_kws=None, **kwargs)

Represents changes in the rankings of the alternatives as lines flowing through the ranking-methods.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• grid_kws (dict or None) – Dict with keyword arguments passed to matplotlib.axes.plt.Axes.grid

• kwargs – Other keyword arguments are passed to the seaborn.lineplot() function. except for data, estimator and sort.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

reg(*, untied=False, r2=True, palette=None, legend=True, r2_fmt='.2g', r2_kws=None, **kwargs)

Plot a pairwise rankings data and a linear regression model fit.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• r2 (bool, default True) – If True, the coefficient of determination results are added to the regression legend.

• palette (matplotlib/seaborn color palette, default None) – Set of colors for mapping the hue variable.

• legend (bool, default True) – If False, suppress the legend for semantic variables.

• r2_fmt (str, default "2.g") – String formatting code to use when adding the coefficient of determination.

• r2_kws (dict or None) – Dict with keywords arguments passed to sklearn.metrics.r2_score() function.

• kwargs – Other keyword arguments are passed to the seaborn.lineplot() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

heatmap(*, untied=False, **kwargs)

Plot the rankings as a color-encoded matrix.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the seaborn.heatmap() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

corr(*, untied=False, corr_kws=None, **kwargs)

Plot the pairwise correlation of rankings as a color-encoded matrix.

By default the pearson correlation coefficient is used.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• corr_kws (dict or None) – Dict with keywords arguments passed the pandas.DataFrame.corr() method.

• kwargs – Other keyword arguments are passed to the seaborn.heatmap() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

cov(*, untied=False, cov_kws=None, **kwargs)

Plot the pairwise covariance of rankings as a color-encoded matrix.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• cov_kws (dict or None) – Dict with keywords arguments passed the pandas.DataFrame.cov() method.

• kwargs – Other keyword arguments are passed to the seaborn.heatmap() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

r2_score(untied=False, r2_kws=None, **kwargs)

Plot the pairwise coefficient of determination regression score function of rankings as a color-encoded matrix.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• cov_kws (dict or None) – Dict with keywords arguments passed the pandas.DataFrame.cov() method.

• kwargs – Other keyword arguments are passed to the seaborn.heatmap() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

distance(*, untied=False, metric='hamming', distance_kws=None, **kwargs)

Plot the pairwise distance between rankings as a color-encoded matrix.

By default the ‘hamming’ distance is used, which is simply the proportion of disagreeing components in Two rankings.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• metric (str or function, default "hamming") – The distance metric to use. The distance function can be ‘braycurtis’, ‘canberra’, ‘chebyshev’, ‘cityblock’, ‘correlation’, ‘cosine’, ‘dice’, ‘euclidean’, ‘hamming’, ‘jaccard’, ‘jensenshannon’, ‘kulczynski1’, ‘mahalanobis’, ‘matching’, ‘minkowski’, ‘rogerstanimoto’, ‘russellrao’, ‘seuclidean’, ‘sokalmichener’, ‘sokalsneath’, ‘sqeuclidean’, ‘yule’.

• distance_kws (dict or None) – Dict with keywords arguments passed the scipy.spatial.distance.pdist function

• kwargs – Other keyword arguments are passed to the seaborn.heatmap() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

box(*, untied=False, **kwargs)

Draw a boxplot to show rankings with respect to alternatives.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the seaborn.boxplot() function.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

bar(*, untied=False, **kwargs)

Draw plot that presents ranking of alternatives by method with vertical bars.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the pandas.Dataframe.plot.bar() method.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

barh(*, untied=False, **kwargs)

Draw plot that presents ranking of alternatives by method with horizontal bars.

Parameters:

• untied (bool, default False) – If it is True and any ranking has ties, the RankResult.untied_rank_ property is used to assign each alternative a single ranked order. On the other hand, if it is False the rankings are used as they are.

• kwargs – Other keyword arguments are passed to the pandas.Dataframe.plot.barh() method.

Return type:

matplotlib.axes.Axes or numpy.ndarray of them

skcriteria.cmp.ranks_cmp.mkrank_cmp(*ranks)

Construct a RankComparator from the given rankings.

This is a shorthand for the RankComparator constructor; it does not require, and does not permit, naming the estimators. Instead, their names will be set to the method attribute of the rankings automatically.

Parameters:

*ranks (list of RankResult objects) – List of the scikit-criteria RankResult objcects.

Returns:

rcmp – Returns a scikit-criteria RanksComparator object.

Return type:

RanksComparator