Adding fix for expected Cython errors - closes #58

.travis.yml

@@ -59,11 +59,11 @@ jobs:
     script: tox -e test-py36
     python: 3.6
     name: Unit Tests
-  - stage: test
-    script: python ci/travis_appveyor_build.py
-    python: 3.6
-    name: Unit Tests (Windows)
-    if: branch in (master, develop) OR tag IS present
+  # - stage: test
+  #   script: python ci/travis_appveyor_build.py
+  #   python: 3.6
+  #   name: Unit Tests (Windows)
+  #   if: branch in (master, develop) OR tag IS present
   - stage: examples-test
     script: tox -e examples-py27
     python: 2.7
@@ -79,7 +79,7 @@ jobs:
   - stage: build
     python: 3.6
     name: Build
-    before_deploy: python ci/get_appveyor_wheels.py
+    # before_deploy: python ci/get_appveyor_wheels.py
     deploy:
     - provider: pypi
       user: "$PYPI_USERNAME"

Pipfile

@@ -9,4 +9,3 @@ scipy = "==1.0.0"
 matplotlib = "==2.1.0"
 
 [dev-packages]
-

appveyor.yml

@@ -25,7 +25,7 @@ environment:
 
 install:
   # We need wheel installed to build wheels
-  - cmd: "%PYTHON%\\python.exe -m pip install wheel tox numpy cython"
+  - cmd: "%PYTHON%\\python.exe -m pip install wheel tox numpy cython twine"
 
 build: off
 
@@ -50,12 +50,14 @@ after_test:
   # 64-bit Python 3.3/3.4. And you need to use %PYTHON% to get the correct
   # interpreter
   - ps: if ($(& $env:PYTHON_EXE .\ci\get_version.py) -ne '') {& $env:PYTHON_EXE setup.py bdist_wheel} ELSE {echo "Not building because this is not a tag build"}
-  - ps: if ($LAST_EXIT_CODE -eq 'True') {ls dist}
+  - ps: if (& $env:PYTHON_EXE .\ci\get_version.py) {ls dist}
+
 artifacts:
   # bdist_wheel puts your built wheel in the dist directory
   - path: dist\*
 
-#on_success:
+on_success:
 #  You can use this step to upload your artifacts to a public website.
 #  See Appveyor's documentation for more details. Or you can simply
-#  access your wheels from the Appveyor "artifacts" tab for your build.
+#  access your wheels from the Appveyor "artifacts" tab for your build.
+  - ps if ($(& $env:PYTHON_EXE .\ci\get_version.py) -ne '') {& $env:PYTHON_EXE setup.py bdist_wheel} ELSE {echo "Not building because this is not a tag build"}

src/MTfit/algorithms/markov_chain_monte_carlo.py

@@ -1516,7 +1516,7 @@ def new_sample(self, jump=0.0, gaussian_jump=False):
                                                                           gaussian_jump=gaussian_jump)
                 return mt
             except Exception:
-                logging.exception('Cython error')
+                logging.exception('WARNING: Error running cython code, resorting to python code')
         else:
             logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
         # Otherwise/Fallback to use python code
@@ -1611,7 +1611,8 @@ def _acceptance_check(self, xi_1, ln_pi_1, scale_factori_1=False):
                     return xi_1, ln_pi_1, scale_factori_1[index], index
                 return xi_1, ln_pi_1, False, index
             except Exception:
-                logger.exception('Cython Error')
+                if not isinstance(xi_1, dict):
+                    logger.exception('WARNING: Error running cython code, resorting to python code')
         else:
             logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
         # Otherwise use/fallback to Python code

src/MTfit/probability/probability.py

@@ -137,7 +137,7 @@ def polarity_ln_pdf(a, mt, sigma, incorrect_polarity_probability=0.0, _use_c=Non
         except Exception as e:
             # Run using python
             # Testing C code
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
             if _C_LIB_TESTS:
                 raise e
     else:
@@ -171,7 +171,7 @@ def polarity_ln_pdf(a, mt, sigma, incorrect_polarity_probability=0.0, _use_c=Non
             ln_p = cprobability.ln_prod(ln_p)
         except Exception:
             if cprobability:
-                logger.exception('Error running cython code')
+                logger.exception('WARNING: Error running cython code, resorting to python code')
             ln_p = np.sum(ln_p, 0)
     if isinstance(ln_p, np.ndarray):
         ln_p[np.isnan(ln_p)] = -np.inf
@@ -255,7 +255,7 @@ def polarity_probability_ln_pdf(a, mt, positive_probability, negative_probabilit
         except Exception as e:
             # Run using python
             # Testing C code
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
             if _C_LIB_TESTS:
                 raise e
     else:
@@ -360,7 +360,7 @@ def amplitude_ratio_ln_pdf(ratio, mt, a_x, a_y, percentage_error_x, percentage_e
         except Exception as e:
             # Run using python
             # Testing C code
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
             if _C_LIB_TESTS:
                 raise e
     else:
@@ -394,7 +394,7 @@ def amplitude_ratio_ln_pdf(ratio, mt, a_x, a_y, percentage_error_x, percentage_e
             ln_p = cprobability.ln_prod(ln_p)
         except Exception:
             if cprobability:
-                logger.exception('Error running cython code')
+                logger.exception('WARNING: Error running cython code, resorting to python code')
             ln_p = np.sum(ln_p, 0)
     if isinstance(ln_p, np.ndarray):
         ln_p[np.isnan(ln_p)] = -np.inf
@@ -479,7 +479,7 @@ def relative_amplitude_ratio_ln_pdf(x_1, x_2, mt_1, mt_2, a_1, a_2, percentage_e
         except Exception as e:
             # Run using python
             # Testing C code
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
             if _C_LIB_TESTS:
                 raise e
     else:
@@ -826,7 +826,7 @@ def dkl(ln_probability_p, ln_probability_q, dV=1.0):
         try:
             return cprobability.dkl(ln_probability_p.copy(), ln_probability_q.copy(), dV)
         except Exception:
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
     else:
         logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
     ind = ln_probability_p > -np.inf
@@ -845,7 +845,7 @@ def dkl(ln_probability_p, ln_probability_q, dV=1.0):
                   ln_probability_q[ind]*probability_p[ind]) * dV
 
 
-def ln_marginalise(ln_pdf, axis=0, dV=1.0):
+def ln_marginalise(ln_pdf, axis=0, dV=1.0, _cprob_err=True):
     """
     Marginalise the pdf from the log pdf input
 
@@ -870,7 +870,8 @@ def ln_marginalise(ln_pdf, axis=0, dV=1.0):
                 return cprobability.ln_marginalise(ln_pdf._ln_pdf.astype(np.float64))
             return cprobability.ln_marginalise(ln_pdf.astype(np.float64))
         except Exception:
-            logger.exception('Error running cython code')
+            if _cprob_err:
+                logger.exception('WARNING: Error running cython code, resorting to python code')
     else:
         logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
     # scale and then marginalise:
@@ -922,7 +923,7 @@ def ln_normalise(ln_pdf, dV=1):
                 normalised_ln_pdf = cprobability.ln_normalise(ln_pdf)
             return normalised_ln_pdf
         except Exception:
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
     else:
         logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
     # scale and then marginalise:
@@ -993,7 +994,7 @@ def dkl_estimate(ln_pdf, V, N):
         try:
             return cprobability.dkl_uniform(ln_pdf.copy(), V, dV)
         except Exception:
-            logger.exception('Error running cython code')
+            logger.exception('WARNING: Error running cython code, resorting to python code')
     else:
         logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
     ind = ln_pdf > -np.inf
@@ -1263,17 +1264,18 @@ def output(self, normalise=True):
             return self.marginalise().normalise()
         return self.marginalise()
 
-    def exp(self):
+    def exp(self, _cprob_err=True):
         if cprobability:
             try:
                 return cprobability.ln_exp(self._ln_pdf)
             except Exception:
-                logger.exception('Error running cython code')
+                if _cprob_err:
+                    logger.exception('WARNING: Error running cython code, resorting to python code')
         else:
             logger.info(C_EXTENSION_FALLBACK_LOG_MSG)
         return np.exp(self._ln_pdf)
 
-    def nonzero(self, discard=100000., n_samples=0):
+    def nonzero(self, discard=100000., n_samples=0, _cprob_err=True):
         """
         Return the non-zero indices of the pdf
 
@@ -1286,7 +1288,7 @@ def nonzero(self, discard=100000., n_samples=0):
             discard: float - discard scale [default = 100000.]
             n_samples: integer - number of samples generated [default = 0]
         """
-        ln_pdf = np.array(self.marginalise(axis=0)._ln_pdf).flatten()
+        ln_pdf = np.array(self.marginalise(axis=0, _cprob_err=_cprob_err)._ln_pdf).flatten()
         m_val = -np.inf
         if n_samples > 0 and discard > 0:
             m_val = max(ln_pdf) - np.log(discard*n_samples)
@@ -1308,7 +1310,7 @@ def normalise(self, dV=False):
         new._ln_pdf = ln_normalise(self._ln_pdf, self.dV)
         return new
 
-    def marginalise(self, axis=0, dV=False):
+    def marginalise(self, axis=0, dV=False, _cprob_err=True):
         """
         Marginalise the pdf object over a given axis
 
@@ -1322,7 +1324,7 @@ def marginalise(self, axis=0, dV=False):
         if dV:
             self._set_dv(dV)
         new = self.__class__(dV=self.dV)
-        new._ln_pdf = ln_marginalise(self._ln_pdf, axis=axis, dV=self.dV)
+        new._ln_pdf = ln_marginalise(self._ln_pdf, axis=axis, dV=self.dV, _cprob_err=_cprob_err)
         return new
 
     def append(self, other, axis=1):

src/MTfit/sampling.py

@@ -170,7 +170,7 @@ def output(self, normalise=True, convert=False, n_samples=0, discard=10000, mcmc
         # Check if there are samples
         if len(ln_pdf):
             # Get non_zero samples
-            non_zero = ln_pdf.nonzero(discard=discard, n_samples=n_samples)
+            non_zero = ln_pdf.nonzero(discard=discard, n_samples=n_samples, _cprob_err=False)
             if discard and n_samples:
                 output_string += 'After discard, '+str(non_zero.shape[0])+' samples remain\n\n'
             if len(ln_pdf.shape) > 1:
@@ -428,7 +428,7 @@ def ln_bayesian_evidence(output, n_samples, prior=_6sphere_prior):
         p = prior(output['g'], output['d'])
     if not isinstance(output['ln_pdf'], LnPDF):
         output['ln_pdf'] = LnPDF(output['ln_pdf'])
-    return np.log((output['ln_pdf']+np.log(p)-output['ln_pdf']._ln_pdf.max()).exp().sum())+output['ln_pdf']._ln_pdf.max()-np.log(n_samples)
+    return np.log((output['ln_pdf']+np.log(p)-output['ln_pdf']._ln_pdf.max()).exp(_cprob_err=False).sum())+output['ln_pdf']._ln_pdf.max()-np.log(n_samples)
 
 
 def _convert(moment_tensors, i=None):