Hey everyone,
I’m looking for a space-efficient solution for saving the block state of a nested partition to use later, but I’m having trouble with the serialization. I tried pickling the blockstate, but in graph-tool 2.68 this leads to an error:
import pickle
from graph_tool.all import *
g = collection.data["celegansneural"]
state = minimize_nested_blockmodel_dl(g)
block_state = state.get_bs()
block_state # This looks fine, I can use it to reconstruct the NestedBlockState object
new_state = NestedBlockState(g, bs=block_state)
new_state.get_bs() # same as before
But if I save this block state to a pickle, I can’t recover it due to the original graph reference.
with open('test_block.pkl', 'wb') as fh:
pickle.dump(block_state, fh)
with open ('test_block.pkl', "rb") as fh:
bs = pickle.load(fh)
# This leads to an error:
new_state_from_pickle = NestedBlockState(g, bs=bs)
I’m pretty sure this used to work when get_bs() returned a list of numpy arrays (as the documentation states), but now it returns a list that starts with a VertexPropertyMap followed by a series of PropertyArray corresponding to the block state. I tried converting these to np.arrays and passing them to NestedBlockState(), but the new state ends up with a different block state.
I suppose I can save the full NestedBlockState object, but I wanted to avoid that because some of the graphs I’m working with are big enough that this would generate large intermediary files.
Thanks!