Make tests

[mhtess]

No description provided.

[mhtess]

We need to test for each of the webppl() arguments.
Each of the inference algorithms (rejection, mcmc, enumerate, incrementalMH,...) should probably be tested with the inference_opts argument. Here is a model we can use as a template

model <- '
var k = 5
var n = 10

var model = function() {
   var p = uniform( {a:0, b:1} )
   factor(Binomial( {p : p, n: n }).score(k))
   return {p: p}
}
Infer({method:"rejection", samples:1000}, model)
'

webppl(model)

Then

webppl(model, 
inference_opts = list(method = rejection, samples = 1000), 
model_var = "model")

Then

var k = data.k
var n = data.n
...

webppl(model, data = list(k=5, n = 10), data_var = data)

and so on.

[hawkrobe]

@mhtess What other tests still need to be written?

[mhtess]

Some ideas (not sure if they're necessary and/or feasible)

• other functions that ship with rwebppl (get_samples, install_webppl(), install_webppl_package(), link_webppl() if that will still exist)
• multiple chains
• program_file argument

[hawkrobe]

Now that we have install_webppl(), we might want to build in some tests that check backwards compatibility with old versions of webppl to a certain extent (e.g. at least through 0.8.0 or whatever it was when we first started using rwebppl)

[mhtess]

Add test for precision in data passing, a la #57

[mhtess]

updated list of tests to make

• test for kill_webppl() (Make way to kill running webppl program #22)
• test for precision in data passing (toJSON rounds by default to 4 decimal places #57)
• multiple chains
• program_file argument
• some backwards compatibility check