Fig. 5: Demonstration of the convergence and effectiveness of CRANE.
From: Counterfactual mobility network embedding reveals prevalent accessibility gaps in U.S. cities
A Convergence with sample size. The percentage of difference with PSM results decreases with the number of sampled random walks on the urban mobility network and is reduced under 10% when the sample size is over 200,000. B 2D visualization of learned embeddings. Four category embeddings are fixed to be projected onto (0,1), (1,0), (0,-1), and (-1,0). Dashed arrows represent the directions pointing to the highest L(T) from the lowest L(T)’s embedding in the projected 2D space of each demographic feature.
