def testDataShuffler2(self):
## Ensure that the data shuffler is reproducible, i.e. that running it
## multiple times given a random_state produces the same results.
random_state_1 = np.random.RandomState(30)
random_state_2 = np.random.RandomState(30)
inputs = np.arange(200)
outputs = np.arange(500, 700)
data_shuffler1 = forgetting_nuisance.DataShuffler(
inputs, outputs, random_state_1)
data_shuffler2 = forgetting_nuisance.DataShuffler(
inputs, outputs, random_state_2)
sample_inputs_1, sample_outputs_1 = data_shuffler1.next_batch(200)
sample_inputs_2, sample_outputs_2 = data_shuffler2.next_batch(200)
self.assertTrue(np.array_equal(sample_inputs_1, sample_inputs_2))
self.assertTrue(np.array_equal(sample_outputs_1, sample_outputs_2))
def setUp(self):
super(ForgettingNuisanceTest, self).setUp()
self.num = 10
## Make some data with controls versus drug.
control = np.random.multivariate_normal([-10, 0], self.C(10, 2, np.pi / 6),
self.num)
drug = np.random.multivariate_normal([-20, 0], self.C(10, 2, np.pi / 6),
self.num)
inputs = np.vstack((control, drug))
outputs = np.zeros((inputs.shape[0], 2))
outputs[:self.num, 0] = 1
outputs[self.num:, 1] = 1
self.random_state = np.random.RandomState(seed=42)
self.shuffler = forgetting_nuisance.DataShuffler(inputs, outputs,
self.random_state)
## Make data with control versus drug on two separate batches.
controls_batch_1 = np.random.multivariate_normal([0, 0], self.C(1, 1, 0),
self.num)
controls_batch_2 = np.random.multivariate_normal([10, 0], self.C(1, 1, 0),
self.num)
drug_batch_1 = controls_batch_1 + [3, 2]
drug_batch_2 = controls_batch_2 + [3, -2]
labels = ([("control", "batch1")] * len(controls_batch_1) + [
("control", "batch2")
] * len(controls_batch_2) + [("drug", "batch1")] * len(drug_batch_1) +
[("drug", "batch2")] * len(drug_batch_1))
self.dummy_df = pd.DataFrame(
np.vstack(
[controls_batch_1, controls_batch_2, drug_batch_1, drug_batch_2]),
index=pd.MultiIndex.from_tuples(labels, names=["compound", "batch"]))
def testDataShuffler3(self):
## Ensure data shuffler works properly when batch size is larger than number
## of input/output pairs.
inputs = np.arange(4)
outputs = np.arange(10, 14)
random_state = np.random.RandomState(30)
data_shuffler = forgetting_nuisance.DataShuffler(
inputs, outputs, random_state)
sample_inputs_1, sample_outputs_1 = data_shuffler.next_batch(200)
sample_inputs_2, sample_outputs_2 = data_shuffler.next_batch(200)
self.assertCountEqual(sample_inputs_1, sample_inputs_2)
self.assertCountEqual(sample_outputs_1, sample_outputs_2)
