Revert back to using tf.distribution instead of tfp

For reasons not fully understood, using TensorFlow Probability causes
Travis CI to use too much memory and a core dump occurs.

So until this is understood all use of tfp should be reverted to a
combination of tf.distributions and
tf.contrib.distributions (for the Poisson).

This is really annoying, but at least a temporary fix that doesn't break
all functionality and gains.

c.f.
- pytest-dev/pytest#3527
- travis-ci/travis-ci#9153

Author: matthewfeickert

.travis.yml

@@ -5,15 +5,13 @@ python:
   - '2.7'
   - '3.6'
 before_install:
-  - pip install --upgrade pip setuptools
+  - pip install --upgrade pip setuptools wheel
 install:
   - pip install --ignore-installed -U -q -e .[tensorflow,torch,mxnet,minuit,develop]  # Ensure right version of NumPy installed
+  - pip freeze
 script:
   - pyflakes pyhf
-  # pytest takes up too much memory in Travis if run all at once, so split off
-  # test_notebooks.py into its own pytest run
   - pytest --ignore tests/benchmarks/ --ignore tests/test_notebooks.py
-  - pytest tests/test_notebooks.py
 after_success: coveralls
 
 # always test (on both 'push' and 'pr' builds in Travis)
@@ -34,21 +32,31 @@ env:
 
 jobs:
   include:
+  -
+    python: '2.7'
+    script:
+      - pytest tests/test_notebooks.py
+  -
+    python: '3.6'
+    script:
+      - pytest tests/test_notebooks.py
   - stage: benchmark
     python: '3.6'
     before_install:
-      - pip install --upgrade pip setuptools
+      - pip install --upgrade pip setuptools wheel
     install:
       - pip install --ignore-installed -U -q -e .[tensorflow,torch,mxnet,develop]
+      - pip freeze
     script: pytest --benchmark-sort=mean tests/benchmarks/
   - stage: docs
     python: '3.6'
     before_install:
       - sudo apt-get update
       - sudo apt-get -qq install pandoc
-      - pip install --upgrade pip setuptools
+      - pip install --upgrade pip setuptools wheel
     install:
       - pip install --ignore-installed -U -q -e .[tensorflow,torch,mxnet,develop]
+      - pip freeze
     script:
     - python -m doctest README.md
     - cd docs && make html && cd -

pyhf/tensor/tensorflow_backend.py

@@ -1,6 +1,6 @@
 import logging
 import tensorflow as tf
-import tensorflow_probability as tfp
+# import tensorflow_probability as tfp
 
 log = logging.getLogger(__name__)
 
@@ -226,7 +226,8 @@ def poisson(self, n, lam):
         """
         n = self.astensor(n)
         lam = self.astensor(lam)
-        return tf.exp(tfp.distributions.Poisson(lam).log_prob(n))
+        # return tf.exp(tfp.distributions.Poisson(lam).log_prob(n))
+        return tf.exp(tf.contrib.distributions.Poisson(lam).log_prob(n))
 
     def normal(self, x, mu, sigma):
         """
@@ -256,7 +257,8 @@ def normal(self, x, mu, sigma):
         x = self.astensor(x)
         mu = self.astensor(mu)
         sigma = self.astensor(sigma)
-        normal = tfp.distributions.Normal(mu, sigma)
+        # normal = tfp.distributions.Normal(mu, sigma)
+        normal = tf.distributions.Normal(mu, sigma)
         return normal.prob(x)
 
     def normal_cdf(self, x, mu=0, sigma=1):
@@ -286,5 +288,6 @@ def normal_cdf(self, x, mu=0, sigma=1):
         x = self.astensor(x)
         mu = self.astensor(mu)
         sigma = self.astensor(sigma)
-        normal = tfp.distributions.Normal(mu, sigma)
+        # normal = tfp.distributions.Normal(mu, sigma)
+        normal = tf.distributions.Normal(mu, sigma)
         return normal.cdf(x)

setup.py

@@ -34,8 +34,8 @@
     ],
     'tensorflow':[
        'tensorflow>=1.10.0',
+       # 'tensorflow-probability>=0.3.0', # Causing troulbe with Travis CI, but *should* be used
        'numpy<=1.14.5,>=1.14.0',  # Lower of 1.14.0 instead of 1.13.3 to ensure doctest pass
-       'tensorflow-probability>=0.3.0',
        'setuptools<=39.1.0',
     ],
     'develop': [