Error in train_for_fig5a_images.py

[WangYuNeng]

Hi,

I try to run train_for_fig5a_images.py and got error at initializingdtm = DTM(cfg).

...
poisson_binomial_ising_graph_manager.py", line 132, in make_base_graph
    assert n_visible_nodes < (
AssertionError: n_visible=5688 > size/2=3200. The code's logic assumes all visible nodes can fit on the 'upper' half of the grid.

Looks like because each pixel is encoded with 7 nodes, the graph requires at least 784 * 7 * 2 = 10976 nodes while the dtm is initialized with 80 * 80= 6400 nodes. Probably because the deprecated version does not have this constraint?

Any fix to reproduce the figure5a image or anyway I can run the old version?