Creating the 3rd-order Model using the "Kenyan Households" dataset fails because the dataset seems to be too large. But the number of first-order edges is only 65 286 which seems to be manageable to me. The Problem is probably that there are only 42 different time stamps leading to a large amount of links happening simultaneously which in turn leads to a huge amount of 2nd and 3rd order links. Many discrete-time graphs probably have similar properties.
Maybe there is still some optimization potential for this specific type of network that we could look into.
Working Example:
https://github.com/jusch2109/pathpyG/blob/a3c0b2d877137c08c96791fe53ac7dd8a2169772/src/pathpyG/visualisations/Project_JS/HotVisHouseholds.ipynb
On CPU, the Kernel crashes and on GPU, there is a memory overflow.
Creating the 3rd-order Model using the "Kenyan Households" dataset fails because the dataset seems to be too large. But the number of first-order edges is only 65 286 which seems to be manageable to me. The Problem is probably that there are only 42 different time stamps leading to a large amount of links happening simultaneously which in turn leads to a huge amount of 2nd and 3rd order links. Many discrete-time graphs probably have similar properties.
Maybe there is still some optimization potential for this specific type of network that we could look into.
Working Example:
https://github.com/jusch2109/pathpyG/blob/a3c0b2d877137c08c96791fe53ac7dd8a2169772/src/pathpyG/visualisations/Project_JS/HotVisHouseholds.ipynb
On CPU, the Kernel crashes and on GPU, there is a memory overflow.