ECR / Docker Image with GConstruct to Partition Graph Issue
- Dominant language
- Python
- Stars
- 452
- Forks
- 76
- PR merge metrics
- No merged PRs in 30d
Description
Hello!
I am trying to setup GraphStorm. Currently I used the Docker Instructions at https://github.com/awslabs/graphstorm/tree/main/docker/sagemaker to setup my images. However I run into an issue with autogluon.
Traceback (most recent call last):
File "", line 189, in _run_module_as_main
File "", line 112, in _get_module_details
File "/usr/local/lib/graphstorm/python/graphstorm/__init__.py", line 30, in
from . import gsf
File "/usr/local/lib/graphstorm/python/graphstorm/gsf.py", line 62, in
from .model.embed import (GSPureLearnableInputLayer,
File "/usr/local/lib/graphstorm/python/graphstorm/model/__init__.py", line 23, in
from .lm_embed import (GSLMNodeEncoderInputLayer,
File "/usr/local/lib/graphstorm/python/graphstorm/model/lm_embed.py", line 35, in
from .utils import (
File "/usr/local/lib/graphstorm/python/graphstorm/model/utils.py", line 38, in
from ..gconstruct.file_io import stream_dist_tensors_to_hdf5
File "/usr/local/lib/graphstorm/python/graphstorm/gconstruct/__init__.py", line 19, in
from .construct_graph import verify_confs
File "/usr/local/lib/graphstorm/python/graphstorm/gconstruct/construct_graph.py", line 38, in
from .transform import parse_feat_ops, process_features, preprocess_features
File "/usr/local/lib/graphstorm/python/graphstorm/gconstruct/transform.py", line 40, in
from autogluon.tabular import TabularPredictor
ModuleNotFoundError: No module named 'autogluon'
When I add this onto the Docker, I eventually run into another
FileNotFoundError: Cannot find DGL C++ graphbolt library at /opt/conda/lib/python3.11/site-packages/dgl/graphbolt/libgraphbolt_pytorch_2.9.1.so
I maybe doing things wrong, but really would appreciate help on this issue.
Contributor guide
Research direction
Start with the docker/sagemaker instructions and reproduce the import failure through graphstorm/gconstruct/transform.py, where autogluon.tabular is loaded. Then inspect the reported DGL GraphBolt library path and compare the image's installed versions. Done means the ECR Docker image can run the GConstruct partition workflow without either missing-dependency error.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- docker, python, pytorch
- Domain
- devops, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 28/100