AttributeError: 'Graph' object has no attribute 'node'

[amitoengg]

I am getting Graph' object has no attribute 'node' error when i call LabelModel.fit on a simple example. Not sure whats wrong. (adding cardinality param doesnt help)

Installation

Ubuntu subsystem on Windows 10 (same error in standalone Ubuntu)
Python 3.6.7
Same error message on 0.9.0 and 0.9.1

pip3 install snorkel==0.9.0

Collecting snorkel==0.9.0
Downloading https://files.pythonhosted.org/packages/42/12/9c5ac9e097574a0657026af7cd127fddbe8f0888dc4fccedb26a8c29e1b6/snorkel-0.9.0-py3-none-any.whl (131kB)
100% |████████████████████████████████| 133kB 666kB/s
Requirement already satisfied: pandas<0.25.0,>=0.24.0 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: scipy<2.0.0,>=1.2.0 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: tensorboardX<2.0,>=1.6 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: torch<1.2.0,>=1.1.0 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: networkx<3.0,>=2.2 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: tqdm<5.0.0,>=4.29.0 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: numpy<2.0.0,>=1.16.0 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: scikit-learn<0.22.0,>=0.20.2 in /usr/local/lib/python3.6/dist-packages (from snorkel==0.9.0)
Requirement already satisfied: python-dateutil>=2.5.0 in /usr/lib/python3/dist-packages (from pandas<0.25.0,>=0.24.0->snorkel==0.9.0)
Requirement already satisfied: pytz>=2011k in /usr/local/lib/python3.6/dist-packages (from pandas<0.25.0,>=0.24.0->snorkel==0.9.0)
Requirement already satisfied: protobuf>=3.8.0 in /usr/local/lib/python3.6/dist-packages (from tensorboardX<2.0,>=1.6->snorkel==0.9.0)
Requirement already satisfied: six in /usr/lib/python3/dist-packages (from tensorboardX<2.0,>=1.6->snorkel==0.9.0)
Requirement already satisfied: decorator>=4.3.0 in /usr/local/lib/python3.6/dist-packages (from networkx<3.0,>=2.2->snorkel==0.9.0)
Requirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-learn<0.22.0,>=0.20.2->snorkel==0.9.0)
Requirement already satisfied: setuptools in /usr/lib/python3/dist-packages (from protobuf>=3.8.0->tensorboardX<2.0,>=1.6->snorkel==0.9.0)
Installing collected packages: snorkel
Successfully installed snorkel-0.9.0

Code

import numpy as np
from snorkel.labeling import LabelModel
L = np.array([[0, 0, -1], [-1, 0, 1], [1, -1, 0]])
Y_dev = [0, 1, 0]
label_model = LabelModel(verbose=False)
label_model.fit(L)

Traceback (most recent call last):
File "/usr/local/lib/python3.6/dist-packages/snorkel/labeling/model/label_model.py", line 743, in fit
self._create_tree()
File "/usr/local/lib/python3.6/dist-packages/snorkel/labeling/model/label_model.py", line 554, in _create_tree
self.c_tree = get_clique_tree(nodes, [])
File "/usr/local/lib/python3.6/dist-packages/snorkel/labeling/model/graph_utils.py", line 49, in get_clique_tree
S = G2.node[i]["members"].intersection(G2.node[j]["members"])
AttributeError: 'Graph' object has no attribute 'node'

[dataframing]

Hi @amitoengg, this is due to a recent issue I noted on #1491: networkx made a backwards-incompatible change to their Graph attributes, combined with snorkel assuming that they wouldn't make backwards-incompatible changes until 3.0. The patch has been merged to master, but in the meantime you can make sure to also specify networkx==2.3.0 in order for the LabelModel to run just fine.

Let me know if this works!

[eggie5]

I'm getting this too starting 2 days ago when my job started failing...

[bhancock8]

@eggie5 are you running from Snorkel from source (in which case, if you pulled from master since #1492 was merged, I'd be surprised and there's more we need to look into) or from pip? If it's from pip, that's because we haven't pushed up the patch yet to PyPI. It should be up soon!

[eggie5]

I'm commenting about the pip release per my comments here 4d579fa#commitcomment-35546869 . Any estimate on the new pip release?

[henryre]

We released Snorkel v0.9.3 last night, which contains this patch. Closing for now, feel free to reopen if this still isn't working.

[Raj9417]

Hi @amitoengg, this is due to a recent issue I noted on #1491: networkx made a backwards-incompatible change to their Graph attributes, combined with snorkel assuming that they wouldn't make backwards-incompatible changes until 3.0. The patch has been merged to master, but in the meantime you can make sure to also specify networkx==2.3.0 in order for the LabelModel to run just fine.

Let me know if this works!

This Solve the problem for me Thanks.

[hmubarak]

hi,
i am suing snorkel 0.9.5 (pip install snorkel) and networkx 2.3 but still have the same problem:
AttributeError: 'Graph' object has no attribute 'node'

I am trying 01_spam_tutorial.ipynb which is part of Snorkel tutorial on GitHub without any modification:
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L_train=L_train, n_epochs=500, log_freq=100, seed=123) <===== Error!

any solution please? thanks

[bhancock8]

@hmubarak could you check in the virtual environment where you're running the tutorial and confirm what version of networkx is present? (e.g., run pip freeze | grep networkx and see a result like networkx==2.4)