Convert continuous and discrete distribution parameters to floatX or int32

[lucianopaz]

This should solve issue #3223. While I was working on PR #3293 I also ran into this problem but decided to submit a separate PR for it. I only explicitly casted to floatX or int32 if the distribution's docstring said that the parameter should be a float or an int. The rest (multivariate, mixture and timeseries) I left unchanged.

[twiecki]

Needs a rebase.

[lucianopaz]

@twiecki, some tests are failing when FLOATX='float32'. Should we mark those as expected failures?

[lucianopaz]

By chance I saw #2366, where it says that int32 should be used if floatX is float64, and that int16 should be used if float32 was set as floatX. Maybe we should do the same here.

[twiecki]

Suggested change

	self.sigma = sigma = tt.as_tensor_variable(floatX(sigma))
	self.sigma = self.sd = sigma = tt.as_tensor_variable(floatX(sigma))

[twiecki]

Suggested change

	r'$\text{sigma} \sim \text{HalfNormal}(\mathit{sd}=1.0)$',
	r'$\text{sigma} \sim \text{HalfNormal}(\mathit{sigma}=1.0)$',

[twiecki]

Suggested change

	r'$\text{alpha} \sim \text{Normal}(\mathit{mu}=0.0,~\mathit{sd}=10.0)$',
	r'$\text{alpha} \sim \text{Normal}(\mathit{mu}=0.0,~\mathit{sigma}=10.0)$',

[twiecki]

Suggested change

	r'$\text{beta} \sim \text{Normal}(\mathit{mu}=0.0,~\mathit{sd}=10.0)$',
	r'$\text{beta} \sim \text{Normal}(\mathit{mu}=0.0,~\mathit{sigma}=10.0)$',

[twiecki]

Suggested change

	r'$\text{Y_obs} \sim \text{Normal}(\mathit{mu}=f(\text{mu}),~\mathit{sd}=f(\text{sigma}))$'
	r'$\text{Y_obs} \sim \text{Normal}(\mathit{mu}=f(\text{mu}),~\mathit{sigma}=f(\text{sigma}))$'

[twiecki]

    def test__repr_latex_(self):
        for distribution, tex in zip(self.distributions, self.expected):
>           assert distribution._repr_latex_() == tex
E           AssertionError: assert '$\\text{Y_ob...ext{sigma}))$' == '$\\text{Y_obs...ext{sigma}))$'
E             - $\text{Y_obs} \sim \text{Normal}(\mathit{mu}=\text{mu},~\mathit{sigma}=f(\text{sigma}))$
E             + $\text{Y_obs} \sim \text{Normal}(\mathit{mu}=f(\text{mu}),~\mathit{sigma}=f(\text{sigma}))$
E             ?                                              ++         +

[twiecki]

Thanks @lucianopaz!