def testPartiallySpecifiedTestSet(self):
"""Check that partially specified test set raises an error."""
num_test_points = 5
dataset = _test_dataset(num_test_points)
del dataset['test_student_ids']
with self.assertRaisesRegex(ValueError, 'all be specified'):
item_response_theory.ItemResponseTheory(**dataset)
def testCreateDataset(self, num_test_points):
"""Checks that creating a dataset works."""
# Technically this is private functionality, but we don't have it tested
# elsewhere.
if not tf.executing_eagerly():
self.skipTest('This is Eager only for now due to _sample_dataset being '
'Eager-only.')
model = item_response_theory.ItemResponseTheory(
**_test_dataset(num_test_points))
model2 = item_response_theory.ItemResponseTheory(
**model._sample_dataset(tfp_test_util.test_seed()))
self.validate_log_prob_and_transforms(
model2,
sample_transformation_shapes=dict(
identity={
'mean_student_ability': [],
'student_ability': [20],
'question_difficulty': [10],
},
test_nll=[],
per_example_test_nll=[num_test_points],
))
def testBasic(self, num_test_points):
"""Checks that you get finite values given unconstrained samples.
We check `unnormalized_log_prob` as well as the values of the sample
transformations.
Args:
num_test_points: Number of test points.
"""
model = item_response_theory.ItemResponseTheory(
**_test_dataset(num_test_points))
self.validate_log_prob_and_transforms(
model,
sample_transformation_shapes=dict(
identity={
'mean_student_ability': [],
'student_ability': [20],
'question_difficulty': [10],
},
test_nll=[],
per_example_test_nll=[num_test_points],
))