def mds(m, axis=0, dim=2, method='spearman', **kwargs):
""" Multidimensional scaling plot from a preference matrix.
See: https://en.wikipedia.org/wiki/Multidimensional_scaling
Args:
m: preference matrix.
dim: number of dimensions to plot (2 or 3).
method: any metric method.
**kwargs: arguments for rk.scatterplot function (e.g. fontsize).
"""
names = None
if axis == 0:
if rk.is_dataframe(m):
names = m.columns
m = m.T # transpose
elif axis == 1:
if rk.is_dataframe(m):
names = m.index
else:
raise Excpetion('axis must be 0 or 1.')
# Compute pairwise distances
dist_matrix = rk.distance_matrix(m, method=method)
# Call the plot functions
mds_from_dist_matrix(dist_matrix, dim=dim, names=names, **kwargs)
def tsne(m, axis=0, dim=2, **kwargs):
""" Use T-SNE algorithm to show the matrix m in a 2 or 3 dimensions space.
Args:
axis: axis of dimensionality reduction.
dim: number of dimensions. 2 for 2D plot, 3 for 3D plot.
**kwargs: arguments for rk.scatterplot function (e.g. fontsize, pointsize).
"""
names = None
if axis == 0:
if rk.is_dataframe(m):
names = m.columns
m = m.T # transpose
elif axis == 1:
if rk.is_dataframe(m):
names = m.index
else:
raise Excpetion('axis must be 0 or 1.')
m_transformed = TSNE(n_components=dim).fit_transform(m)
# Display
scatterplot(m_transformed, dim=dim, names=names, **kwargs)
def show(m,
rotation=90,
title=None,
size=2,
annot=False,
round=2,
color='royalblue',
cmap=None):
""" Display a ranking or a prefrence matrix.
If m is 1D: show ballot (bar plot).
If m is 2D: show preferences (heatmap).
TODO: annot argument adding the values in the plot.
Args:
rotation: x labels rotation.
title: string - title of the figure.
size: integer - higher value for a smaller figure.
annot: If True, write the values.
round: Number of decimals to display if annot is True.
color: Color for 1D bar plot.
cmap: Color map for 2D heatmap.
"""
dim = len(m.shape)
if dim == 1: # 1D
x = np.arange(len(m))
bar_plot = plt.bar(x, m, align='center', color=color)
if annot:
autolabel(bar_plot, m, round=round)
if rk.is_series(m):
plt.xticks(x, m.index, rotation=rotation)
elif dim == 2: # 2D
fig, ax = plt.subplots(figsize=(m.shape[1] / size, m.shape[0] / size))
sns.heatmap(m,
ax=ax,
annot=annot,
linewidths=.2,
fmt='0.' + str(round),
cmap=cmap)
x = np.arange(m.shape[1])
if rk.is_dataframe(m):
plt.xticks(x, m.columns, rotation=rotation)
else:
raise (Exception(
'Passed array must have only 1 or 2 dimension, not {}.'.format(
dim)))
if title is not None:
plt.title(title)
plt.show()
def mds_from_dist_matrix(distance_matrix, dim=2, names=None, **kwargs):
""" Multidimensional scaling plot from a symmetric distance matrix (pairwise distances).
See: https://en.wikipedia.org/wiki/Multidimensional_scaling
Args:
m: distance matrix.
dim: number of dimensions to plot (2 or 3).
names: names of objects. Will be overwritten if distance_matrix is a pd.DataFrame.
**kwargs: arguments for rk.scatterplot function (e.g. fontsize).
"""
if rk.is_dataframe(distance_matrix):
names = distance_matrix.columns
transformer = MDS(n_components=dim, dissimilarity='precomputed')
m_transformed = transformer.fit_transform(distance_matrix)
# Display
scatterplot(m_transformed, dim=dim, names=names, **kwargs)