Skip to main content
. 2020 May 30;2020:298–307.

Figure 2:

Figure 2:

Comparing performance of different methods on training diagnosis models with labels restricted to diseases in Semigran. Error bars over 5 different random sampling of the additional disorders being added, while keeping the random seed fixed. We found similar trends with the neural network model that uses embedding, and do not show results due to space constraints. Error bars computed on 5 different random choices of the additional diseases added.