Thiết kế website giá rẻ

Question

I’m trying to analyze a large graph data using RAPIDS minimum spanning tree package. When I run my code it says RuntimeError. The RAPIDS and CUDA are the latest versions.

The graph is undirected and about 27GB big. Using NVIDIA A100 80GB PCIe (4 of them)

My code:

### Convert KG2 into undirected graph

# Define column names
cols = ['subject', 'object', 'predicate']

# Load the TSV file into a cuDF DataFrame
gdf = cudf.read_csv('graph.tsv', names=cols, delimiter='t')

# Convert node identifiers to strings (if they are not already)
gdf['subject'] = gdf['subject'].astype(str)
gdf['object'] = gdf['object'].astype(str)

# Create a cuGraph Graph object
G = cugraph.Graph()

# Create the graph from the cuDF DataFrame's edgelist
G.from_cudf_edgelist(gdf, source='subject', destination='object')

# List of vertexes 
lst_nodes = ['C:7045767', 'G:0007271', 'G:0035249', 'G:0005947', 'G:0004129', 
    'C:26519', 'U:C0132173', 'G:0018393', 'G:0006979', 'H:0025464', 
    'G:0007613', 'C:64645', 'P:000010173', 'G:0014055', 'G:0061535', 
    'G:0150076', 'M:0001152', 'H:0002185', 'M:0004975', 'E:0005816', 
    'C:60425']

# Check for existing nodes in the graph
existing_nodes = G.nodes().to_pandas().tolist()
missing_nodes = set(lst_nodes) - set(existing_nodes)

# Print missing nodes
if missing_nodes:
    print(f"Some nodes in lst_nodes are not in the graph: {missing_nodes}")
else:
    print("All nodes are present in the graph.")

# Proceed with the rest of the code only if there are no missing nodes
def compute_subgraph_mst(lst_nodes):
    # Compute the MST of the entire graph
    mst = cugraph.minimum_spanning_tree(G)

    # Filter the MST to include only the subgraph with lst_nodes
    subgraph = mst[mst['src'].isin(lst_nodes) | mst['dst'].isin(lst_nodes)]

    # Convert the subgraph to an edge list
    edges = subgraph[['src', 'dst']].to_pandas().values.tolist()

    # Create a new graph from the subgraph
    subgraph_g = cugraph.Graph()
    subgraph_g.from_pandas_edgelist(pd.DataFrame(edges, columns=['src', 'dst']))

    # Compute the MST of the subgraph
    subgraph_mst = cugraph.minimum_spanning_tree(subgraph_g)
    
    return subgraph_mst

# Proceed with the rest of the code only with existing nodes
if not missing_nodes:
    subgraph_mst = compute_subgraph_mst(lst_nodes)
else:
    lst_nodes = [node for node in lst_nodes if node not in missing_nodes]
    print(f"Proceeding with nodes: {lst_nodes}")
    subgraph_mst = compute_subgraph_mst(lst_nodes)

Output:

Some nodes in lst_nodes are not in the graph: {'E:0005816', 'C:60425'}

Proceeding with nodes: ['P:7045767', 'G:0007271', 'G:0035249', 'G:0005947', 'G:0004129', 'C:26519', 'U:C0132173', 'G:0018393', 'G:0006979', 'H:0025464', 'G:0007613', 'C:64645', 'P:000010173', 'G:0014055', 'G:0061535', 'G:0150076', 'M:0001152', 'H:0002185', 'M:0004975']

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[3], line 64
     62 lst_nodes = [node for node in lst_nodes if node not in missing_nodes]
     63 print(f"Proceeding with nodes: {lst_nodes}")
---> 64 subgraph_mst = compute_subgraph_mst(lst_nodes)

Cell In[3], line 38, in compute_subgraph_mst(lst_nodes)
     36 def compute_subgraph_mst(lst_nodes):
     37     # Compute the MST of the entire graph
---> 38     mst = cugraph.minimum_spanning_tree(G)
     40     # Filter the MST to include only the subgraph with lst_nodes
     41     subgraph = mst[mst['src'].isin(lst_nodes) | mst['dst'].isin(lst_nodes)]

File /scratch/USERNAME/anaconda3/envs/rapids-24.04/lib/python3.11/site-packages/cugraph/tree/minimum_spanning_tree.py:105, in minimum_spanning_tree(G, weight, algorithm, ignore_nan)
    103     return cugraph_to_nx(mst)
    104 else:
--> 105     return _minimum_spanning_tree_subgraph(G)

File /scratch/USERNAME/anaconda3/envs/rapids-24.04/lib/python3.11/site-packages/cugraph/tree/minimum_spanning_tree.py:26, in _minimum_spanning_tree_subgraph(G)
     24 if G.is_directed():
     25     raise ValueError("input graph must be undirected")
---> 26 mst_df = minimum_spanning_tree_wrapper.minimum_spanning_tree(G)
     27 if G.renumbered:
     28     mst_df = G.unrenumber(mst_df, "src")

File minimum_spanning_tree_wrapper.pyx:71, in cugraph.tree.minimum_spanning_tree_wrapper.minimum_spanning_tree()

File minimum_spanning_tree_wrapper.pyx:52, in cugraph.tree.minimum_spanning_tree_wrapper.mst_double()

RuntimeError: parallel_for: failed to synchronize: cudaErrorIllegalAddress: an illegal memory access was encountered

Danh mục