Bug fixes for ExactInference on continuous factors

[linzhp]

Variable Elimination for continuous variables was broken when there is only one query variable, as illustrated by the added unit test in this PR. This PR fixed it.

@ankurankan Please review.

[ankurankan]

@linzhp The tests are failing, could you check please ?

[linzhp]

I tried Python 3.6, 3.5 and 2.7 on Windows, but I can't reproduce the test failure. I will try running it on Linux.

Also, from the Travis' log, I saw this:

Setting environment variables from .travis.yml
$ export DISTRIB="conda"
$ export PYTHON_VERSION="3.5"
$ export COVERAGE="true"
$ source ~/virtualenv/python2.7/bin/activate
$ python --version
Python 2.7.13

The Python version it used was wrong. So essentially, Travis only ran the test in Python 2.7 for three times.

[ankurankan]

Yeah, it's not failing on windows. The appveyor builds on windows and it's passing.

The Python version it used was wrong. So essentially, Travis only ran the test in Python 2.7 for three times.

No, the python versions are correct actually. The python --version is ran before we setup our conda environment, so it's basically the default python version on the system. I will actually change this logging, it's a bit confusing.

[codecov]

Codecov Report

Merging #941 into dev will increase coverage by 0.05%.
The diff coverage is 90.47%.

@@            Coverage Diff             @@
##              dev     #941      +/-   ##
==========================================
+ Coverage   94.71%   94.76%   +0.05%     
==========================================
  Files         114      114              
  Lines       11185    11201      +16     
==========================================
+ Hits        10594    10615      +21     
+ Misses        591      586       -5

Impacted Files	Coverage Δ
pgmpy/factors/continuous/ContinuousFactor.py	91.78% <0%> (+5.47%)	⬆️
pgmpy/inference/ExactInference.py	94.81% <100%> (ø)	⬆️
pgmpy/tests/test_inference/test_ExactInference.py	100% <100%> (ø)	⬆️
pgmpy/factors/distributions/CustomDistribution.py	96.72% <90%> (+1.1%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5953dcd...556f87a. Read the comment docs.

[linzhp]

All tests passed now. Please take another look, @ankurankan