def test_seed(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
def get_sampler(seed):
return samp.SeqSampler(dataset, true_frac=0.5, base_seed=seed)
base = 0
sampler = get_sampler(base)
epoch_one = list(sample for sample in iter(sampler))
sampler = get_sampler(base)
epoch_two = list(sample for sample in iter(sampler))
self.assertEqual(epoch_one, epoch_two)
new_sampler = get_sampler(None)
self.assertNotEqual(new_sampler.base_seed, None)
# covering random edge case with probability 2**-32. lel
while new_sampler.base_seed == 0:
new_sampler = get_sampler(None)
new_epoch_one = list(sample for sample in iter(new_sampler))
self.assertNotEqual(epoch_one, new_epoch_one)
def test_getitem(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
value = ((0, 0, slice(0, 20)), (0, 2, slice(10, 30)))
res = dataset[value]
self.assertEqual(len(res), 3)
self.assertEqual(res[0].shape, (2, 20, 3, 120, 120))
self.assertTrue(np.array_equal(res[1], np.array((20, 20))))
self.assertEqual(res[2][0], res[2][1])
value = ((0, 0, slice(0, 20)), (1, 0, slice(10, 30)))
res = dataset[value]
self.assertNotEqual(res[2][0], res[2][1])
with self.assertRaises(ValueError):
dataset[(0, np.arange(90))]
with self.assertRaises(ValueError):
dataset[0]
with self.assertRaises(ValueError):
wrong_value = ((1, slice(0, 20)), (0, slice(20, 40)))
dataset[wrong_value + wrong_value]
def test_size(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
sampler = samp.SeqSampler(dataset, true_frac=0.0)
self.assertTrue(len(sampler), int(np.floor(len(dataset) / 2)))
sampler = samp.SeqSampler(dataset, true_frac=1.0)
self.assertTrue(len(sampler), len(dataset))
epoch = list(sample for sample in iter(sampler))
for itx in range(len(epoch)):
with self.subTest(i=itx):
self.assertEqual(epoch[itx][0][0], epoch[itx][1][0])
sampler = samp.SeqSampler(dataset, true_frac=0.0)
self.assertTrue(len(sampler), int(np.floor(len(dataset))))
epoch = list(sample for sample in iter(sampler))
for itx in range(len(epoch)):
with self.subTest(i=itx):
self.assertNotEqual(epoch[itx][0][0], epoch[itx][1][0])
self.assertEqual(samp.SeqSampler.get_max_size(10, 1.0), 10)
self.assertEqual(samp.SeqSampler.get_max_size(10, 0.0), 5)
def test_repetition(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
base = 0
sampler = samp.RepeatingSeqSampler(dataset, base_seed=base)
sampled_list = [
list(sample for sample in iter(sampler)) for i in range(1, 11)
]
for itx in range(len(sampled_list) - 1):
with self.subTest(i=itx):
self.assertEqual(*sampled_list[itx:itx + 2])
new_sampler = samp.RepeatingSeqSampler(dataset, base_seed=base)
new_sampled = list(sample for sample in iter(new_sampler))
self.assertEqual(new_sampled, sampled_list[0])
new_base = base + 1
diff_sampler = samp.RepeatingSeqSampler(dataset, base_seed=new_base)
diff_sampled = list(sample for sample in iter(diff_sampler))
self.assertNotEqual(new_sampled, diff_sampled)
def test_restore(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
base = 0
from_one = samp.SeqSampler(dataset, start_epoch=1, base_seed=base)
from_one_epochs = [
list(sample for sample in iter(from_one)) for i in range(1, 11)
]
from_five = samp.SeqSampler(dataset, start_epoch=6, base_seed=base)
from_five_epochs = [
list(sample for sample in iter(from_five)) for i in range(6, 11)
]
self.assertEqual(from_one_epochs[5:], from_five_epochs)
new_base = base + 1
new_from_five = samp.SeqSampler(dataset,
start_epoch=6,
base_seed=new_base)
new_from_five_epochs = [
list(sample for sample in iter(new_from_five))
for i in range(6, 11)
]
self.assertNotEqual(new_from_five_epochs, from_five_epochs)
def test_iter(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
sampler = samp.SeqSampler(dataset, true_frac=0.5)
sample_it = iter(sampler)
for sample in sample_it:
self.assertTrue(dataset._valid_t(sample[0]))
self.assertTrue(dataset._valid_t(sample[1]))
def test_make_sample(self):
dataset = data.TrainSeqDataSet(TEST_DATASET)
dataset.pad = 15
sampler = samp.SeqSampler(dataset, true_frac=0.5)
d_iter = (i for i in range(len(dataset)))
rnd = np.random.RandomState(0)
# checking correct (two different seq from the same obj) sample
sample = sampler.make_sample(1, d_iter, rnd)
self.assertTrue(dataset._valid_t(sample[0]))
self.assertTrue(dataset._valid_t(sample[1]))
self.assertEqual(sample[0][0], sample[1][0])
self.assertNotEqual(sample[0][1], sample[1][1])
# checking wrong sample
sample = sampler.make_sample(0, d_iter, rnd)
self.assertTrue(dataset._valid_t(sample[0]))
self.assertTrue(dataset._valid_t(sample[1]))
self.assertNotEqual(sample[0][0], sample[1][0])