Issues with isolates in do-calculus

[sa-and]

Hello,

In some cases when querying on a network on which a do-intervention has been done, an endless-loop in the build_join_tree() function of bbn.py occurs. This seems to happen when the preceding intervention disconnects the graph. Below you can see an example graph on which the problem occurred:

An intervention on x0 should remove the edge from x1 to x0. When querying for x1, the program runs in an endless-loop. More specifically (with bn having the structure as in the image and already fit to data).

ie = InferenceEngine(bn)
ie.do_intervention('x0', 1)
prediction = ie.query()['x1']  # this never terminates without raising any error

Interestingly, doing the intervention ie.do_intervention('x2', 1) on the above graph raises ValueError: Do calculus cannot be applied because it would result in an isolate. While this is already easier to debug, to the best of my knowledge this should also be a valid operation in the setting of causal graphs and do-calculus.

Environment:

python 3.8.5
# Name                    Version                   Build  Channel
backcall                  0.2.0              pyhd3eb1b0_0
ca-certificates           2021.1.19            haa95532_0
causalnex                 0.9.1                    pypi_0    pypi
certifi                   2020.12.5        py38haa95532_0
cloudpickle               1.6.0                    pypi_0    pypi
colorama                  0.4.4              pyhd3eb1b0_0
cycler                    0.10.0                   pypi_0    pypi
decorator                 4.4.2              pyhd3eb1b0_0
future                    0.18.2                   pypi_0    pypi
gym                       0.18.0                   pypi_0    pypi
ipython                   7.20.0           py38hd4e2768_1
ipython_genutils          0.2.0              pyhd3eb1b0_1
jedi                      0.17.0                   py38_0
joblib                    1.0.1                    pypi_0    pypi
kiwisolver                1.3.1                    pypi_0    pypi
matplotlib                3.3.4                    pypi_0    pypi
mypy                      0.812              pyhd3eb1b0_0
mypy_extensions           0.4.3                    py38_0
networkx                  2.5                      pypi_0    pypi
numpy                     1.20.1                   pypi_0    pypi
openssl                   1.1.1j               h2bbff1b_0
pandas                    1.2.2                    pypi_0    pypi
parso                     0.8.1              pyhd3eb1b0_0
patsy                     0.5.1                    pypi_0    pypi
pgmpy                     0.1.13                   pypi_0    pypi
pickleshare               0.7.5           pyhd3eb1b0_1003
pillow                    7.2.0                    pypi_0    pypi
pip                       21.0.1           py38haa95532_0
prettytable               0.7.2                    pypi_0    pypi
prompt-toolkit            3.0.8                      py_0
psutil                    5.8.0            py38h2bbff1b_1
pyglet                    1.5.0                    pypi_0    pypi
pygments                  2.8.0              pyhd3eb1b0_0
pyparsing                 2.4.7                    pypi_0    pypi
python                    3.8.5                h5fd99cc_1
python-dateutil           2.8.1                    pypi_0    pypi
pytz                      2021.1                   pypi_0    pypi
scikit-learn              0.22.2                   pypi_0    pypi
scipy                     1.5.4                    pypi_0    pypi
setuptools                52.0.0           py38haa95532_0
six                       1.15.0                   pypi_0    pypi
sqlite                    3.33.0               h2a8f88b_0
stable-baselines3         0.10.0                   pypi_0    pypi
statsmodels               0.12.2                   pypi_0    pypi
torch                     1.7.1                    pypi_0    pypi
tqdm                      4.57.0                   pypi_0    pypi
traitlets                 5.0.5              pyhd3eb1b0_0
typed-ast                 1.4.2            py38h2bbff1b_1
typing_extensions         3.7.4.3            pyha847dfd_0
vc                        14.2                 h21ff451_1
vs2015_runtime            14.27.29016          h5e58377_2
wcwidth                   0.2.5                      py_0
wheel                     0.36.2             pyhd3eb1b0_0
wincertstore              0.2                      py38_0
wrapt                     1.11.2                   pypi_0    pypi
zlib                      1.2.11               h62dcd97_4

[ukm5]

Thanks for posting this. I think you have summarized what I wanted to.
In fact, I have been facing a similar issue when I try to do a double intervention where I do not get an error saying there is an isolate but an infinite computational query is generated.

[FrancescaSogaroQB]

Thank you @ukm5 for reporting this. This is also linked to issue #45. The above has been fixed with this commit, and will be available in the next CausalNex release.