error in convert_graph_int_to_adj_mat

[pckennethma]

Hello,

I am trying to use Causal Disocvery RL on bnlearn benchmarks. I encounter the error in convert_graph_int_to_adj_mat function.

The input to this function is

[-1903318164   235405414   101482606   495790951   201853294   378349935
 -1634426101 -1718146065   134742090          64   134742086   134742084
   446475107   470616428 -1785775892 -1768316434   201884524   134217728
   201949548 -1903613075   470286702   101187694 -1734505621   503843118
 -2070074547   134217838   513518542   503875886   235405386   445754223
           0      524358   236432367   134742086   134217792   134217792
   503908622]

And the error message follows:

Traceback (most recent call last):
  File "main.py", line 337, in <module>
    main()
  File "main.py", line 285, in main
    graph_batch = convert_graph_int_to_adj_mat(graph_int)
  File "/home/user/Causal_Discovery_RL/src/helpers/analyze_utils.py", line 156, in convert_graph_int_to_adj_mat
    for curr_int in graph_int], dtype=int)
  File "/home/user/Causal_Discovery_RL/src/helpers/analyze_utils.py", line 156, in <listcomp>
    for curr_int in graph_int], dtype=int)
ValueError: invalid literal for int() with base 10: '-'

[pckennethma]

the issue is fixed by changing np.int32 to int in several places of the codebase. the error is due to the overflow of int32.

[pckennethma]

The error would occur again when the node size exceeds 64 (which is the precision limit of int long).

[shaido987]

There is an np.int64 used here which could be the reason for the error. You can try to change it to int similarly to how the np.int32 were changed.

[pckennethma]

Thanks for your prompt reply. It looks like the URL does not point to the codebase of Causal Discovery with RL. Nevertheless, I have already changed the place in the codebase of Causal Discovery with RL. But it looks like a limitation for numpy to process such lengthy int variable.

Traceback (most recent call last):
  File "main.py", line 344, in <module>
    main()
  File "main.py", line 280, in main
    graph_int, score_min, cyc_min = np.array(ls_kv[0][0], dtype=np.int), ls_kv[0][1][1], ls_kv[0][1][-1]
OverflowError: Python int too large to convert to C long

[pckennethma]

I guess changing np.int to np.longlong may be a workaround. but how could I work with larger graphs like andes?

[pckennethma]

It turns out that changing the precision cannot deal with the issue when the node size is higher than 64.

Traceback (most recent call last):
  File "main.py", line 344, in <module>
    main()
  File "main.py", line 280, in main
    graph_int, score_min, cyc_min = np.array(ls_kv[0][0], dtype=np.longlong), ls_kv[0][1][1], ls_kv[0][1][-1]
OverflowError: int too big to convert

[zhushyu]

It turns out that changing the precision cannot deal with the issue when the node size is higher than 64.

Traceback (most recent call last):
  File "main.py", line 344, in <module>
    main()
  File "main.py", line 280, in main
    graph_int, score_min, cyc_min = np.array(ls_kv[0][0], dtype=np.longlong), ls_kv[0][1][1], ls_kv[0][1][-1]
OverflowError: int too big to convert

Sorry for a late reply. A quick fix is to use string, instead of int type, as key. E.g., 000111 is converted to ‘000111‘. We have change this part in internal codes but seemed not to update the github ..

Again, sorry for this.