Monte-Carlo likelihoods (GPflow#799)

* add @markvdw's stochastic likelihood, including the softmax

* initial MC likelihood

* remove MonteCarloLikelihood base class from tests

* fix test

* var of predict_mean_and_var and predict_density for MC likelihood

* factor out MC sampling

* add comment for variance bias

* add tests

* fixes

* use same integration as for GH quadrature in MonteCarloLikelihood.predict_mean_and_var()

* .

* increase rtol

* move to proper use of super()

* move MC integration to quadrature module, similar to ndiagquad

* seed to make test deterministic

* add Assert for shape of Y

* tidy up studentT likelihood

* fix for heteroskedastic likelihoods -- requires logp to always call the Y argument Y

* fix doc

* add assert and equivalence tests for SoftMax

* remove erroneously added file

* rename "probit" to inv_probit (which is what it actually is)

* add assert for num_classes to SoftMax

* fix whitespace

* Update RELEASE.md

* Update RELEASE.md

RELEASE.md

@@ -1,3 +1,7 @@
+# Master branch
+ - Added likelihoods where expectations are evaluated with Monte Carlo (`MonteCarloLikelihood`) (#799).
+ - Added `SoftMax` likelihood (#799).
+
 # Release 1.1
  - Added inter-domain inducing features. Inducing points are used by default and are now set with `model.feature.Z`.