Phil 6.17.20

Listened to a fantastic interview with Nell Irvin Painter (White Supremacy at Home and Abroad):

GPT-2 Agents

  • Working on finding the connections between nodes
  • Now that I know how to add weights to edges, I think I want to add the piece that made the move. It needs to be a list, since multiple types of pieces can connect two squares. Added a dict_array per edge:
    if target not in nlist:
        self.G.add_edge(source, target, weight=0)
        self.G[target]['dict_array'] = []
    self.G[target]['weight'] += 1
    for key, val in data_dict.items():
        a:List = self.G[target]['dict_array']
  • I also realize that moves that repeatedly connect squares are more likely to be close, simply because the available squares of more distant moves increase in a geometric fashion. I added a method that writes out moves to Excel where I can play with them. Here are some moves:


  • In looking at these moves, it does seem to be that the majority of the moves seem to be short (e.g. b6-b7, b6-a7, b6-b5). The only exception is the knight (b6-d7). So I think there is a confidence value that I can calculate for the ‘physical’ adjacency of nodes in a network. This could also apply to belief spaces as well. Most consensus requires coordination and common orientation (pos, heading, speed), so commonly connected topics can be said to be ‘closer’
  • Good chat with Aaron about CVPR and algorithms


  • Finish revisions and send to T and Aaron for review. Last thing is to tie back to ground vehicles in the discussion. Done! I think… Need to read the whole thing and see if it still hangs together
  • 2:00 – Meeting