paper/_make_figures.py

"""Generate paper figures from the published dataset."""
import sys
import pandas as pd
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import numpy as np

OUT_DIR = 'C:/Users/davim/Downloads/dataset/published_dataset/paper/figures'

# === Heatmap macrotema × tipo ===
ct = pd.read_parquet(
    'C:/Users/davim/Downloads/dataset/cross_modal_validation/macrotema_x_tipo_entidade__contagem.parquet',
    engine='fastparquet'
).set_index('macrotheme_label')
if 'URL_DOMINIO' in ct.columns:
    ct = ct.drop(columns=['URL_DOMINIO'])
top_types = ct.sum(axis=0).sort_values(ascending=False).head(12).index.tolist()
ct = ct[top_types]
top15 = ct.sum(axis=1).sort_values(ascending=False).head(15).index
ct = ct.loc[top15]
ct_pct = ct.div(ct.sum(axis=1), axis=0) * 100

fig, ax = plt.subplots(figsize=(10, 7))
im = ax.imshow(ct_pct.values, cmap='YlOrRd', aspect='auto')
ax.set_xticks(range(len(ct_pct.columns)))
ax.set_xticklabels(ct_pct.columns, rotation=45, ha='right', fontsize=9)
ax.set_yticks(range(len(ct_pct.index)))
ax.set_yticklabels(ct_pct.index, fontsize=9)
for i in range(ct_pct.shape[0]):
    for j in range(ct_pct.shape[1]):
        v = ct_pct.values[i, j]
        if v >= 5:
            ax.text(j, i, f'{v:.0f}', ha='center', va='center',
                    fontsize=7, color='black' if v < 25 else 'white')
cbar = fig.colorbar(im, ax=ax, fraction=0.04)
cbar.set_label('% das entidades nao-URL no macrotema', fontsize=9)
ax.set_title('Assinatura semantica por macrotema (top 15 x top 12 tipos)', fontsize=10)
plt.tight_layout()
plt.savefig(f'{OUT_DIR}/heatmap_macrotema_tipo.pdf', bbox_inches='tight')
plt.savefig(f'{OUT_DIR}/heatmap_macrotema_tipo.png', dpi=150, bbox_inches='tight')
plt.close()
print('OK heatmap')

# === Timeline ===
notes = pd.read_parquet(
    'C:/Users/davim/Downloads/dataset/published_dataset/notes_pt.parquet',
    engine='fastparquet'
)
notes['ano_mes'] = pd.to_datetime(notes['created_at']).dt.to_period('M').dt.to_timestamp()
g = notes.groupby(['ano_mes','consenso']).size().unstack(fill_value=0)
for c in ['NMR','CRH','CRNH','Outro']:
    if c not in g.columns:
        g[c] = 0
g = g[['NMR','CRH','CRNH','Outro']]

fig, ax = plt.subplots(figsize=(10, 4))
ax.fill_between(g.index, 0, g['NMR'], color='#cccccc', label='NMR (necessita mais avaliacoes)')
ax.fill_between(g.index, g['NMR'], g['NMR']+g['CRH'], color='#4E79A7', label='CRH (publicada)')
ax.fill_between(g.index, g['NMR']+g['CRH'], g['NMR']+g['CRH']+g['CRNH'], color='#E15759', label='CRNH (nao-util)')
ax.fill_between(g.index, g['NMR']+g['CRH']+g['CRNH'], g.sum(axis=1), color='#F28E2B', label='Outro')
ax.set_xlabel('Mes')
ax.set_ylabel('Notas (PT)')
ax.set_title('Volume mensal de notas em PT por consenso comunitario')
ax.legend(loc='upper left', fontsize=8)
plt.tight_layout()
plt.savefig(f'{OUT_DIR}/timeline_consenso.pdf', bbox_inches='tight')
plt.savefig(f'{OUT_DIR}/timeline_consenso.png', dpi=150, bbox_inches='tight')
plt.close()
print('OK timeline')

# === MF scatter ===
sub = notes.dropna(subset=['coreNoteIntercept','coreNoteFactor1']).sample(
    min(10000, len(notes.dropna(subset=['coreNoteIntercept','coreNoteFactor1']))),
    random_state=42
)
fig, ax = plt.subplots(figsize=(7, 6))
colors = sub['consenso'].map({'CRH':'#4E79A7','NMR':'#cccccc','CRNH':'#E15759','Outro':'#F28E2B'}).fillna('#aaaaaa')
ax.scatter(sub['coreNoteFactor1'], sub['coreNoteIntercept'], c=colors.values, s=4, alpha=0.4)
ax.axhline(0.4, color='black', linestyle='--', alpha=0.3, label='intercept = 0.4 (limiar CRH)')
ax.set_xlabel('coreNoteFactor1 (eixo de polarizacao)')
ax.set_ylabel('coreNoteIntercept (utilidade absoluta)')
ax.set_title('Espaco de pontuacao MF (amostra de 10k notas)')
ax.legend(fontsize=8, loc='upper left')
plt.tight_layout()
plt.savefig(f'{OUT_DIR}/mf_scatter.pdf', bbox_inches='tight')
plt.savefig(f'{OUT_DIR}/mf_scatter.png', dpi=150, bbox_inches='tight')
plt.close()
print('OK mf_scatter')

print('\nFiguras geradas em:', OUT_DIR)