Fig. 3.

P-R classifier performance with $\mathrm{3,151}$ known IO accounts (9) and language-specific training data: combined English and French (red curves$\hspace{0.17em}$, $N=\mathrm{17,999}$), English only (green curves$\hspace{0.17em}$, $N=\mathrm{13,155}$), and French only (blue curves$\hspace{0.17em}$, $N=\mathrm{13,159}$), cross-validated using a $90:10$ split and all data (solid curves) and Snorkel positives omitted (dashed curves).