def main():
df = read_plot_data(overwrite=False)
df = df.reset_index()
for centroid_day, grp in df.groupby('centroid_day'):
df.loc[grp.index, 'h_index'] = grp.h_index.rank(method='first',
ascending=False)
df.loc[grp.index,
'num_retweets'] = grp.num_retweets.rank(method='first',
ascending=False)
df = df.groupby(['centroid_day', 'super_community']).mean().reset_index()
palette = [
'#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b',
'#e377c2', '#7f7f7f', '#bcbd22', '#17becf'
]
color_dict = {
'International expert': palette[4],
'National expert': palette[1],
'Political': palette[2],
'Other': '.5'
}
fig, axes = plt.subplots(1, 2, figsize=(3.3, 1.4))
lw = .8
ms = 4
for sc, grp in df.groupby('super_community'):
axes[0].plot(grp.centroid_day.values,
grp.h_index,
c=color_dict[sc],
label=sc,
ms=ms,
lw=lw,
marker='.')
axes[1].plot(grp.centroid_day.values,
grp.num_retweets,
c=color_dict[sc],
label=sc,
ms=ms,
lw=lw,
marker='.')
_min = 600
_max = 3200
axes[0].set_ylim((_max, _min))
axes[1].set_ylim((_max, _min))
axes[0].set_ylabel('Avg. rank h-index $r_h$')
axes[1].set_ylabel('Avg. rank retweets $r_{rt}$')
for ax in axes:
ax.grid()
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b'))
ax.tick_params(axis='x', direction='out', which='minor', size=2)
ax.tick_params(axis='x', direction='out', which='major', size=2)
ax.tick_params(axis='y', which='major', direction='out', size=2)
ax.locator_params(nbins=4)
sns.despine()
fig.tight_layout()
save_fig(fig, 'fig4cd', 1, dpi=600, plot_formats=['png', 'pdf', 'svg'])
def fig4a():
do_log = False
palette = [
'#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b',
'#e377c2', '#7f7f7f', '#bcbd22', '#17becf'
]
color_dict = {
'International expert': palette[4],
'National expert': palette[1],
'Political': palette[2],
'Other': '.5'
}
df = read_data(overwrite=False)
g = sns.jointplot(data=df,
kind='scatter',
x='h_index_rank',
y='num_retweets_rank',
alpha=.8,
s=4,
ec=None,
hue='super_community',
palette=color_dict.values(),
hue_order=color_dict.keys(),
height=2,
marginal_ticks=False,
space=.1)
num_ranks = len(df) + 50
if do_log:
g.ax_joint.set_xlim((num_ranks, 1))
g.ax_joint.set_ylim((num_ranks, 1))
else:
g.ax_joint.set_xlim((num_ranks, -30))
g.ax_joint.set_ylim((num_ranks, -30))
g.ax_joint.grid()
g.ax_joint.get_legend().remove()
g.ax_joint.set_xlabel(r'Rank $h$-index $r_h$')
g.ax_joint.set_ylabel('Rank retweets $r_{rt}$')
g.ax_marg_x.tick_params(axis='x', direction='out', which='major', size=0)
g.ax_marg_y.tick_params(axis='y', direction='out', which='major', size=0)
g.ax_marg_y.set_xlim((None, g.ax_marg_y.get_xlim()[1] * 1.1))
g.ax_marg_x.set_ylim((None, g.ax_marg_x.get_ylim()[1] * 1.1))
# log scale
if do_log:
g.ax_joint.set_xscale('log')
g.ax_joint.set_yscale('log')
# diagonal line
g.ax_joint.plot([0, num_ranks], [0, num_ranks],
color='0.15',
ls='dashed',
lw=.5)
# num ticks
num_ticks = 5
g.ax_joint.xaxis.set_major_locator(plt.MaxNLocator(num_ticks))
g.ax_joint.yaxis.set_major_locator(plt.MaxNLocator(num_ticks))
save_fig(g, 'fig4a', 1, dpi=600, plot_formats=['png', 'pdf', 'svg'])
def main():
df = get_category_data(without_other=False)
df_entropy = get_geo_entropy()
df = df.T
df = df.divide(df.sum(axis=1), axis=0)
df.columns = df.columns.tolist()
df['Internationality'] = df_entropy['Internationality']
df = df[[c for c in df.columns if c != 'Other']]
max_val = df.subtract(df.mean(axis=0), axis=1).divide(df.std(axis=0),
axis=1).max().max()
usecols = [
'Political Supporter', 'Public Services', 'Politics & Government',
'Media', 'Healthcare', 'Science', 'Internationality'
]
df = df[usecols]
width = 2.2
height = 3.7
# cmap = None
cmap = 'RdBu_r'
# cmap = 'Blues'
# cmap = 'Reds'
palette = [c['color'] for c in matplotlib.rcParams['axes.prop_cycle']]
colors = {
'International expert': palette[4],
'National elites': palette[1],
'Political': palette[2],
'Other': '.5'
}
#df.index.name = 'communities'
super_communities = [
community_to_types[com] for com in df.index
] #df.reset_index().communities.apply(lambda s: community_to_types[s]).copy()
super_community_colors = [colors[sc] for sc in super_communities
] #super_communities.apply(lambda s: colors[s])
df.to_csv('df_cat.csv')
X = df.values
Z = scipy.stats.zscore(X, axis=0)
Z[Z < .5] = 0
f = plt.figure(dpi=300)
sns.heatmap(Z, annot=True, xticklabels=usecols, yticklabels=alphab[:15])
plt.savefig('../plots/Zmatrix.png')
print(Z)
for i, com in enumerate(list(df.index)):
for j, cat in enumerate(usecols):
if Z[i, j] > 0:
print('Com {} exceed of {}'.format(com, cat))
Xpca = PCA().fit_transform(Z)
fig, axes23 = plt.subplots(2, 3, figsize=(6, 4))
for method, axes in zip(['average', 'ward'], axes23):
z = hierarchy.linkage(Z, method=method)
# Plotting
axes[0].plot(range(1, len(z) + 1), z[::-1, 2], label='distance')
knee = np.diff(z[::-1, 2], 2)
axes[0].plot(range(2, len(z)), knee, label='2nd deriv of distance')
axes[0].legend(fontsize=4)
num_clust1 = knee.argmax() + 2
knee[knee.argmax()] = 0
num_clust2 = knee.argmax() + 2
print(num_clust1)
print(num_clust2)
axes[0].text(num_clust1, z[::-1, 2][num_clust1 - 1],
'possible\n<- knee point')
part1 = hierarchy.fcluster(z, num_clust1, 'maxclust')
part2 = hierarchy.fcluster(z, num_clust2, 'maxclust')
clr = [
'#2200CC', '#D9007E', '#FF6600', '#FFCC00', '#ACE600', '#0099CC',
'#8900CC', '#FF0000', '#FF9900', '#FFFF00', '#00CC01', '#0055CC',
'red'
]
for part, ax in zip([part1, part2], axes[1:]):
for cluster in set(part):
ax.scatter(Xpca[part == cluster, -2],
Xpca[part == cluster, -1],
color=clr[cluster])
for i in range(Xpca.shape[0]):
ax.annotate(alphab[i], (Xpca[i, -2], Xpca[i, -1]))
m = '\n(method: {})'.format(method)
plt.setp(axes[0],
title='Screeplot{}'.format(m),
xlabel='partition',
ylabel='{}\ncluster distance'.format(m))
plt.setp(axes[1], title='{} Clusters'.format(num_clust1))
plt.setp(axes[2], title='{} Clusters'.format(num_clust2))
plt.tight_layout()
plt.savefig('../plots/linkage.png')
row_linkage = hierarchy.linkage(Z, method='average', metric='euclidean')
#print(row_linkage)
#print(df.index)
clus = hierarchy.fcluster(row_linkage, t=1.1, depth=2)
print('Nclus: ', len(set(clus)), clus)
clus = hierarchy.fcluster(row_linkage, t=1.2, depth=3)
print('Nclus: ', len(set(clus)), clus)
clus = hierarchy.fcluster(row_linkage, t=3.1, criterion='distance')
print('Nclus: ', len(set(clus)), clus)
clus = hierarchy.fcluster(row_linkage, t=3.3, criterion='distance')
print('Nclus: ', len(set(clus)), clus)
print(df.index)
g = sns.clustermap(
df, #.reset_index(drop=True),
method='ward',
dendrogram_ratio=.3,
colors_ratio=.04,
cmap=cmap,
center=0,
lw=0.4,
col_cluster=False,
z_score=1,
figsize=(width, height),
cbar_pos=(1.1, .33, .02, .2),
row_colors=super_community_colors)
fig = plt.gcf()
fig.delaxes(g.ax_col_dendrogram)
g.ax_row_colors.get_xaxis().set_visible(False)
# colorbar formatting
g.ax_cbar.set_ylabel('Z-score', rotation=0, ha='left', va='center')
g.ax_cbar.locator_params(nbins=5)
g.ax_cbar.tick_params(length=2,
axis='both',
which='major',
direction='out')
# create type legend
type_legend_patches = [
mpatches.Patch(color=c, label=l) for l, c in colors.items()
]
labels = list(colors.keys())
#labels[labels.index('National expert')] = 'National elite'
legend_type = g.ax_heatmap.legend(labels=labels,
loc='center left',
bbox_to_anchor=(1.3, .85),
handles=type_legend_patches,
frameon=False,
title='Super-community',
handlelength=.6,
handleheight=1)
# move axis labels
g.ax_heatmap.set_xticklabels(g.ax_heatmap.get_xticklabels(),
rotation=75,
ha='right')
g.ax_heatmap.set_yticklabels(g.ax_heatmap.get_yticklabels(), rotation=0)
#g.ax_heatmap.set_yticklabels(g.ax_heatmap.get_yticklabels(), rotation=0)
g.ax_row_colors.set_xticklabels(g.ax_row_colors.get_xticklabels(),
rotation=75,
ha='right')
offset = matplotlib.transforms.ScaledTranslation(.05, 0,
fig.dpi_scale_trans)
for ax in [g.ax_heatmap, g.ax_row_colors]:
for label in ax.xaxis.get_majorticklabels():
label.set_transform(label.get_transform() + offset)
# other cosmetics
g.ax_heatmap.tick_params(axis='both',
which='both',
bottom=False,
right=False)
g.ax_row_colors.tick_params(axis='both',
which='both',
bottom=False,
right=False)
g.ax_heatmap.set_ylabel('Community')
save_fig(fig, 'fig2', 1, plot_formats=['png', 'pdf'], dpi=600)
def fig4b():
palette = [
'#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', '#8c564b',
'#e377c2', '#7f7f7f', '#bcbd22', '#17becf'
]
color_dict = {
'International expert': palette[4],
'National expert': palette[1],
'Political': palette[2],
'Other': '.5'
}
df = read_data()
fig, ax = plt.subplots(1, 1, figsize=(2, 2), sharex=False, sharey=False)
for _t, grp in df.groupby('super_community'):
means = grp.mean()
ci = bootstrap_ci(grp, ['h_index_rank', 'num_retweets_rank'])
for k, v in ci.items():
ci[k] = [[abs(means[k] - v[0])], [abs(means[k] - v[1])]]
ax.errorbar(means['h_index_rank'],
means['num_retweets_rank'],
xerr=ci['h_index_rank'],
yerr=ci['num_retweets_rank'],
label=_t,
c=color_dict[_t],
ms=5,
marker='.',
capsize=.8,
elinewidth=.5,
markeredgewidth=.5)
num_ranks = len(df)
ax.set_xlim((num_ranks, 0))
ax.set_ylim((num_ranks, 0))
ax.grid(True)
# num ticks
num_ticks = 4
ax.xaxis.set_major_locator(plt.MaxNLocator(num_ticks))
ax.yaxis.set_major_locator(plt.MaxNLocator(num_ticks))
# annotate
len_arr = 500
center = num_ranks / 2
arrow_props = dict(fc='.15',
shrink=.1,
width=.15,
headwidth=2,
headlength=2)
ax.annotate('',
xy=(center - len_arr, center + len_arr),
xytext=(center, center),
ha='left',
va='top',
arrowprops=arrow_props)
ax.annotate('',
xy=(center + len_arr, center - len_arr),
xytext=(center, center),
ha='left',
va='top',
arrowprops=arrow_props)
ax.text(center + len_arr - 200,
center - len_arr,
'Over-ranked\nby retweets\n(high virality)',
ha='right',
va='bottom',
fontsize=6)
ax.text(center - len_arr,
center + len_arr,
'Under-ranked\nby retweets\n(low virality)',
ha='left',
va='top',
fontsize=6)
# axis labels
ax.set_xlabel(r'Avg. rank $h$-index $r_h$')
ax.set_ylabel('Avg. rank retweets $r_{rt}$')
sns.despine()
ax.set_aspect('equal')
# diagonal line
ax.plot([0, num_ranks], [0, num_ranks],
color='0.15',
ls='dashed',
lw=.5,
zorder=0)
save_fig(fig, 'fig4b', 2, dpi=600, plot_formats=['png', 'pdf', 'svg'])
def main():
# read data
df_int = get_interaction_data()
df_size = get_community_size_data()
df_size = df_size[df_int.index.min():df_int.index.max()]
df_size['Total'] = df_size.sum(axis=1)
dfs = {}
for _type in ['inter_cluster', 'intra_cluster', 'received']:
df = df_int[df_int.columns[df_int.columns.str.startswith(_type)]].copy()
col_rename = {}
for col in df.columns:
col_rename[col] = col.split('_')[-1]
df = df.rename(columns=col_rename)
df['Total'] = df.sum(axis=1)
dfs[_type] = df
col_order = ['Total', 'International expert', 'National expert', 'Political', 'Other']
palette = [c['color'] for c in matplotlib.rcParams['axes.prop_cycle']]
# orange, blue,
color_dict = {'International expert': palette[4], 'National expert': palette[1], 'Political': palette[2], 'Other': '.5', 'Total': '0.15'}
ls_dict = {'International expert': '-', 'National expert': '-', 'Political': '-', 'Other': '-', 'Total': 'dotted'}
m_dict = {'International expert': '.', 'National expert': '.', 'Political': '.', 'Other': '.', 'Total': None}
z_order = {'International expert': 4, 'National expert': 3, 'Political': 2, 'Other': 1, 'Total': 0}
colors = [color_dict[c] for c in col_order]
col_order_rev = col_order[::-1]
colors_rev = [color_dict[c] for c in col_order_rev]
# plot
fig, all_axes = plt.subplots(2, 2, figsize=(4.5, 3.5), sharex=False, sharey=False)
# specs
lw = 1
ec = 'white'
ms = 5
# panels A
ax = all_axes[0][0]
for col in df[col_order]:
ax.plot(df_size.index.values, df_size[col].values, label=col, color=color_dict[col], marker=m_dict[col], ms=ms, lw=lw, ls=ls_dict[col], zorder=z_order[col])
ax.ticklabel_format(axis="y", style="sci", scilimits=(0, 0))
ax.set_ylabel('$N$')
ax.set_title('Number of users', fontsize=7)
ax.set_ylim((0, 0.6*10**7))
ax.grid()
# panels B
ax = all_axes[0][1]
df = dfs['received']
df /= df_size # normalize by size of type
for col in df[col_order]:
ax.plot(df.index.values, df[col].values, label=col, color=color_dict[col], marker=m_dict[col], ms=ms, lw=lw, ls=ls_dict[col], zorder=z_order[col])
ax.set_ylabel('$A_u$')
ax.set_title('Avg. attention per user', fontsize=7)
ax.set_ylim((0, 8))
ax.grid()
# panels C
df = dfs['inter_cluster']
ax = all_axes[1][0]
for col in df[col_order]:
ax.plot(df.index.values, df[col], label=col, color=color_dict[col], marker=m_dict[col], ms=ms, lw=lw, ls=ls_dict[col], zorder=z_order[col])
ax.set_title('External attention component', fontsize=7)
ax.set_ylabel(r'$a^{ext}$')
ax.set_xlabel('')
ax.grid()
ax.set_ylim((0, .35))
# panels D
df = dfs['intra_cluster']
ax = all_axes[1][1]
for col in df[col_order]:
ax.plot(df.index.values, df[col], label=col, color=color_dict[col], marker=m_dict[col], ms=ms, lw=lw, ls=ls_dict[col], zorder=z_order[col])
ax.set_title('Internal attention component', fontsize=7)
ax.set_ylabel(r'$a^{int}$')
ax.set_xlabel('')
ax.grid()
ax.set_ylim((0, .95))
# vertical lines
for i, time_pos in enumerate([2, 9]):
for ax_row in all_axes:
for ax in ax_row:
ax.axvline(df.index[time_pos], ls='dashed', lw=.5, c='.15', zorder=0)
ax.annotate(chr(97 + i), (df.index[time_pos], ax.get_ylim()[1]*.95), xytext=(2, 0), textcoords='offset points', ha='left', va='center')
# tick style
for ax_row in all_axes:
for ax in ax_row:
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b'))
ax.tick_params(axis='x', direction='out', which='minor', size=2)
ax.tick_params(axis='x', direction='out', which='major', size=2)
ax.tick_params(axis='y', which='major', direction='out', size=2)
sns.despine()
fig.tight_layout()
fig.subplots_adjust(hspace=.5)
save_fig(plt.gcf(), 'fig3', 1, dpi=600, plot_formats=['png', 'pdf'])