DistributionOfStancePerDomain.py

from matplotlib import pyplot as plt
from matplotlib import ticker as mtick
import seaborn as sns
import pandas as pd
import numpy as np

sns.set_theme(style='whitegrid')

dfs = [pd.read_json(r'../../Data/Model Predicted Data/%s-Predictions.json' % ds) for ds in ['NLP', 'ML']]
dfs = [df[['url', 'stance']].drop_duplicates() for df in dfs]

cp = sns.color_palette()[:8] + sns.color_palette('muted')[:8]

# distribution of stance per domain (Histplot and Bar)
fig, (ax1, ax2) = plt.subplots(ncols=2, gridspec_kw={'width_ratios': [8, 2], 'wspace': 0.05}, figsize=(5.4, 4.0))

dfbar = pd.DataFrame({'Negative': [0.0, 0.0], 'Positive': [0.0, 0.0]}, index=['NLP', 'ML'])
for k, domain in enumerate(['NLP', 'ML']):
	# copy
	dft = dfs[k].copy(deep=True)
	
	# lineplot
	sns.histplot(data=dft, x='stance', element='poly', color=cp[k*3], fill=False, stat='probability', binrange=(-1.05, 1.05), bins=20, ax=ax1, label=domain)
	
	# count classes
	dft['stance'] = np.where(dft['stance'] < 0, 'Negative', 'Positive')
	dft['venue'] = 'Class'
	dft = dft.groupby(['stance', 'venue']).size().reset_index().pivot(columns='stance', index='venue', values=0)
	dft = dft[['Negative', 'Positive']]
	dft = dft.div(dft.sum(1), axis=0).mul(100)
	dfbar['Negative'][domain] = dft['Negative']
	dfbar['Positive'][domain] = dft['Positive']

dfbar.plot(kind='bar', stacked=True, rot=0, color=[cp[3], cp[0]], width=0.65, legend=False, ax=ax2)
ax1.yaxis.set_major_formatter(mtick.FuncFormatter(lambda x,_: str('%f'%(100*x)).rstrip('0').rstrip('.')+'%'))
ax2.yaxis.set_major_formatter(mtick.PercentFormatter())
ax2.tick_params(labelleft=False, labelright=True)

plt.subplots_adjust(top=0.93)
plt.suptitle('Distribution of Stance per Domain', fontsize='medium')
ax1.set_xlabel('Stance')
ax2.set_xlabel('Class')
ax1.set_ylabel('Percentage of Papers')
ax2.set_ylabel(None)
plt.subplots_adjust(bottom=0.14, left=0.15, right=0.89)
plt.savefig('DistributionOfStancePerDomain.pdf')
plt.show()