def test_de_noise(self):
with tf.Graph().as_default():
counts = [[1, 2, 3], [1, 2, 3]]
noise = [[3, 3, 4], [3, 2, 1]]
with tf.compat.v1.Session() as sess:
# Larger noise ratio -> the results are more sharp.
de_noised = sess.run(utils.de_noise(counts, noise, ratio=0.9))
self.assertAllClose(de_noised, [[0., 0.22, 0.78], [0., 0., 1.]])
# Smaller noise ratio -> the results are close to counts.
de_noised = sess.run(utils.de_noise(counts, noise, ratio=0.1))
self.assertAllClose(de_noised, [
[0.151852, 0.337037, 0.511111],
[0.12963, 0.333333, 0.537037],
])
def test_de_noise_exception(self):
with tf.Graph().as_default():
with self.assertRaises(ValueError):
utils.de_noise([[1, 2, 3]], [[1, 2, 3]], ratio=1.1)
with self.assertRaises(ValueError):
utils.de_noise([[1, 2, 3]], [[1, 2, 3]], ratio=-0.1)
with self.assertRaises(tf.errors.InvalidArgumentError):
utils.de_noise([[-1, 2, 3]], [[1, 2, 3]])
with self.assertRaises(tf.errors.InvalidArgumentError):
utils.de_noise([[1, 2, 3]], [[0, 2, 3]])