def test_recursive_passed():
target_f1 = Mock()
target_g1 = Mock()
lookup = {
'f1': target_f1,
'g1': target_g1}
spec = {
'function': 'f1',
'parameters': {
'a': {'value': 2},
'b': {
'generator': 'g1',
'parameters': {
'c': {'expose': 'outer_c'}
}
}
}
}
func = evaluate_distribution(spec, lookup.__getitem__)
del lookup['f1']
del lookup['g1']
del spec['function']
del spec['parameters']
assert func._exposed_kwargs == {'outer_c'}
result = func('rstate', outer_c=1)
target_g1.assert_called_once_with('rstate', c=1)
target_f1.assert_called_once_with(a=2, b=target_g1())
assert result == target_f1()
with pytest.raises(TypeError):
func('rstate')
def test_value():
''' Only one use case here. All arguments ignored and
a constant is returned. '''
spec = {'value': 'result'}
func = evaluate_distribution(spec, None)
del spec['value']
assert func._exposed_kwargs == set()
assert func('rstate') == 'result'
assert func('rstate', a=1, b=1) == 'result'
def test_generator_noarg():
target_gen = Mock()
generator_lookup = {'return': target_gen}
spec = {'generator': 'return'}
func = evaluate_distribution(spec, generator_lookup.__getitem__)
del generator_lookup['return']
del spec['generator']
assert func._exposed_kwargs == set()
result = func('rstate')
target_gen.assert_called_once_with('rstate')
assert result == target_gen()
def test_declare():
target_f1 = Mock()
target_f2 = Mock()
target_f3 = Mock()
function_lookup = {
'f1': target_f1,
'f2': target_f2,
'f3': target_f3
}
spec = {
'function': 'f1',
'parameters': {
'a': {
'function': 'f2',
'parameters': {
'b': {'value': 1},
'c': {'expose': 'arg1'}
}
}
},
'declare': {
'arg1': {
'function': 'f3',
'parameters': {
'd': {'expose': 'arg2'}
}
}
}
}
func = evaluate_distribution(spec, function_lookup.__getitem__)
del function_lookup['f1']
del function_lookup['f2']
del function_lookup['f3']
assert func._exposed_kwargs == {'arg2'}
result = func('rstate', arg2=2)
target_f3.assert_called_once_with(d=2)
target_f2.assert_called_once_with(b=1, c=target_f3())
target_f1.assert_called_once_with(a=target_f2())
assert result == target_f1()
target_f1.reset_mock()
target_f2.reset_mock()
target_f3.reset_mock()
# declarations override any input arguments
result = func('rstate', arg1=4, arg2=2)
target_f3.assert_called_once_with(d=2)
target_f2.assert_called_once_with(b=1, c=target_f3())
target_f1.assert_called_once_with(a=target_f2())
assert result == target_f1()
def test_eval_order():
''' Tests the order of evaluation w.r.t. the random number generator,
which directly impacts the result.
The defined order of evaluation at each level is:
1. declarations, in sorted name order
2. parameters, in sorted name order
3. target function
'''
target_g1 = Mock(side_effect=lambda r: r())
def mock_func(rstate, **kwargs):
return rstate()
def returning(arg):
return arg
target_g2 = Mock(side_effect=mock_func)
lookup = {
'return': returning,
'g1': target_g1,
'g2': target_g2
}
spec = {
'generator': 'g2',
'declare': {
'arg1': {'generator': 'g1'},
'arg2': {'generator': 'g1'}
},
'parameters': {
'a': {'generator': 'g1'},
'b': {'generator': 'g1'},
'c': {'generator': 'g1'},
'd': {
'function': 'return',
'parameters': {
'arg': {'expose': 'arg1'}
}
},
'e': {
'function': 'return',
'parameters': {
'arg': {'expose': 'arg2'}
}
}
}
}
func = evaluate_distribution(spec, lookup.__getitem__)
rstate = Mock(side_effect=[1, 2, 3, 4, 5, 6])
result = func(rstate)
target_g2.assert_called_once_with(rstate, d=1, e=2, a=3, b=4, c=5)
assert result == 6
def test_noarg_function():
''' function is just a special case of generator, where
rstate is ignored. This is implemented using a wrapper
function, so further tests can just be performed for the
generator spec. '''
target_func = Mock()
function_lookup = {'return': target_func}
spec = {'function': 'return'}
func = evaluate_distribution(spec, function_lookup.__getitem__)
del function_lookup['return']
del spec['function']
assert func._exposed_kwargs == set()
result = func('rstate')
target_func.assert_called_once_with()
assert result == target_func()
def test_generator_recursive():
target_gen = Mock()
generator_lookup = {'return': target_gen}
spec = {
'generator': 'return',
'parameters': {
'a': {'value': 1}
}
}
func = evaluate_distribution(spec, generator_lookup.__getitem__)
del generator_lookup['return']
del spec['generator']
del spec['parameters']
assert func._exposed_kwargs == set()
result = func('rstate')
target_gen.assert_called_once_with('rstate', a=1)
assert result == target_gen()
def test_generator_expose():
target_gen = Mock()
generator_lookup = {'return': target_gen}
spec = {
'generator': 'return',
'parameters': {
'a': {'expose': 'b'}
}
}
func = evaluate_distribution(spec, generator_lookup.__getitem__)
del generator_lookup['return']
del spec['generator']
del spec['parameters']
assert func._exposed_kwargs == {'b'}
result = func('rstate', b=2)
target_gen.assert_called_once_with('rstate', a=2)
assert result == target_gen()
with pytest.raises(TypeError):
func('rstate')
def get_sample_function(specification):
''' Still working on this spec... read a random generator specification
(see example files) including evaluation/attribute commands. Returns a
function which, given a random seed, generates an instance and returns
its attributes. '''
# Generating function taking a randgen object and returning an instance.
generate = evaluate_distribution(specification['instances'], lookup)
# Function to evaluate features of the generated instance.
attributes = lookup(specification['attributes'])
# Sample function returns a data point given a seed value.
def sample(seed):
randgen = random.Random(seed)
instance = generate(randgen)
return dict(attributes(instance), seed=seed)
return sample