Figure 5.

Example for the importance of stratification. For this example, we use a threshold of 2.5 mmol CO₂/g to group structures in low and high performing materials, which is slightly higher than the threshold chosen by Boyd et al.¹³ Then, we randomly draw 100 structures and can observe that the class distribution gets distorted—sometimes we do not have any high performing materials in our sample. Stratification can be used to remedy this effect.