def _test_case(scene, expression, expected, msg=None):
from pprint import pprint
print("Objects:")
pprint(scene['objects'])
model = Model(scene, ontology)
expr = Expression.fromstring(expression)
value = model.evaluate(expr)
print(expr)
print("Expected:", expected)
print("Observed:", value)
eq_(value, expected, msg)
def _test_case(parser, sentence, lf, scene=None):
model = Model(scene or {"objects": []}, ontology)
results = parser.parse(sentence.split())
for result in results:
semantics = result.label()[0].semantics()
print("Predicted semantics:", semantics)
if str(semantics) == lf:
printCCGDerivation(result)
return True
ok_(
False,
"Expected successful parse with correct lf. Found %i parses; none matched LF."
% len(results))
def ecTaskAsPyCCGUpdate(task, ontology):
"""
Converts an EC task into a PyCCG update.
Assumes the task has a single scene and the instructions as a string.
:return:
tokenized_instruction, model, goal.
"""
remove_punctuation = str.maketrans('', '', string.punctuation)
tokenized = task.features.translate(remove_punctuation).lower().split()
scene, goal = task.examples[0] # Unpack the examples.
scene, goal = scene[0], goal
return tokenized, Model(scene, ontology), goal
def _make_mock_model(learner):
"""
Mock learner does not have any grounding -- just build a spurious Model
instance.
"""
return Model({"objects": []}, learner.ontology)
def iter_data():
"""Helper function for iterating dataset and inject the ontology to make the scene a model for execution."""
for v, q, a in dataset:
yield (Model(v, ontology), q.split(), a)
""",
ontology,
include_semantics=True)
#######
# Execute on a scene.
scene = {
"objects": [
Object("sphere", "big", "rubber"),
Object("cube", "small", "metal"),
Object("cylinder", "small", "rubber"),
]
}
model = Model(scene, ontology)
print("the ball")
print(model.evaluate(Expression.fromstring(r"unique(\x.has_shape(x,sphere))")))
######
# Parse an utterance and execute.
learner = WordLearner(lex)
# Update with distant supervision.
learner.update_with_distant("the cube".split(), model, scene['objects'][1])
parser = learner.make_parser()
results = parser.parse("the cube".split())
printCCGDerivation(results[0])
learner = WordLearner(initial_lex)
i = 100
from pprint import pprint
for idx in sorted_idxs:
_, objects_i, instruction_i, goal_i = dataset[idx]
goal_i = tuple(goal_i)
if instruction_i != "one above triangle":
continue
print(i)
print(instruction_i)
scene = process_scene(objects_i)
model = Model(scene, ontology)
print(np.array(objects_i))
print(goal_i)
instruction_i = instruction_i.split()
results = learner.update_with_distant(instruction_i, model, goal_i)
if results:
printCCGDerivation(results[0][0])
assert False
print("Lexicon:")
learner.lexicon.debug_print()
print()