def test_classwise_covariance_2class_batches_float32(self):
from tensorflow_gan.python.eval import eval_utils
"""Checks value consistency of streaming_covariance."""
if tf.executing_eagerly():
# tf.placeholder() is not compatible with eager execution.
return
np.random.seed(0)
data = [[1., 2.], [2., 1.], [2., 4.], [4., 2.]]
labels = [0,0, 1, 1]
placeholder = tf.compat.v1.placeholder(dtype=tf.float64, shape=(2,2))
placeholder_labs = tf.compat.v1.placeholder(dtype=tf.int32, shape=(2))
value, update_op = eval_utils.streaming_classwise_autocovariance(x=placeholder, labels=placeholder_labs, nclass=2)
expected_result0 = np.cov(data[:2], rowvar=False)
expected_result1 = np.cov(data[2:], rowvar=False)
expected_result = np.stack([expected_result0, expected_result1])
with self.cached_session() as sess:
sess.run(tf.compat.v1.initializers.local_variables())
# batch 1/2
update_op_result = sess.run(update_op, feed_dict={placeholder: data[:2], placeholder_labs: labels[:2]})
result = sess.run(value)
self.assertAllClose(update_op_result, result)
# batch 2/2
update_op_result = sess.run(update_op, feed_dict={placeholder: data[2:], placeholder_labs: labels[2:]})
result = sess.run(value)
self.assertAllClose(update_op_result, result)
# overall
self.assertAllClose(expected_result, result)
def test_classwise_covariance_2class_random_singlet_batches_float32(self):
from tensorflow_gan.python.eval import eval_utils
"""Checks value consistency of streaming_covariance."""
if tf.executing_eagerly():
# tf.placeholder() is not compatible with eager execution.
return
np.random.seed(0)
n_egs_per_class = 8
data = np.random.randn(n_egs_per_class*2,2)
labels = ([0]*n_egs_per_class) + ([1]*n_egs_per_class)
placeholder = tf.compat.v1.placeholder(dtype=tf.float64, shape=(1,2))
placeholder_labs = tf.compat.v1.placeholder(dtype=tf.int32, shape=(1,))
value, update_op = eval_utils.streaming_classwise_autocovariance(x=placeholder, labels=placeholder_labs, nclass=2)
expected_result0 = np.cov(data[:n_egs_per_class], rowvar=False)
expected_result1 = np.cov(data[n_egs_per_class:], rowvar=False)
expected_result = np.stack([expected_result0, expected_result1])
with self.cached_session() as sess:
sess.run(tf.compat.v1.initializers.local_variables())
for i in range(n_egs_per_class*2):
update_op_result = sess.run(update_op, feed_dict={placeholder: [data[i]], placeholder_labs: [labels[i]]})
result = sess.run(value)
self.assertAllClose(update_op_result, result)
# overall
self.assertAllClose(expected_result, result)
def test_classwise_covariance_1class_simple_float32(self):
from tensorflow_gan.python.eval import eval_utils
"""Checks handling of float32 values in test_classwise_covariance_float32."""
if tf.executing_eagerly():
# streaming_covariance is not supported when eager execution is enabled.
return
x = [[1., 2.], [2., 1.]]
result, update_op = eval_utils.streaming_classwise_autocovariance(
x=tf.constant(x, dtype=tf.float32), labels=tf.constant([0,0]), nclass=1)
expected_result = np.expand_dims(np.cov(x, rowvar=False), axis=0)
with self.cached_session() as sess:
sess.run(tf.compat.v1.initializers.local_variables())
self.assertAllClose(expected_result, update_op)
self.assertAllClose(expected_result, result)