def test_matmul(self):
# computation
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(3, 4))
c = tf.matmul(a, b)
# value
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
# test
self.run(c, tf_feed_dict={a: a_val, b: b_val})
def test_binary_ops(self):
# computation
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(2, 3))
c = tf.add(a, b)
d = tf.mul(c, a)
e = tf.div(d, b)
f = tf.sub(a, e)
g = tf.maximum(a, f)
# value
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
# test
self.run(g, tf_feed_dict={a: a_val, b: b_val})
def test_neg(self):
# computation
a = tf.placeholder(tf.float32, shape=(20, 30))
neg_a = tf.neg(a)
# test
feed_dict = {a: np.random.rand(*tf_to_shape_tuple(a))}
self.run(neg_a, tf_feed_dict=feed_dict)
def test_relu(self):
# computation
a = tf.placeholder(tf.float32, shape=(100, 200))
f = tf.nn.relu(a)
# value
feed_dict = {a: np.random.randn(*tf_to_shape_tuple(a))}
# test
self.run(f, tf_feed_dict=feed_dict)
def test_identity(self):
# computation
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(2, 3))
c = tf.identity(a) + b
f = tf.identity(c)
# value
feed_dict = dict()
for x in [a, b]:
feed_dict[x] = np.random.rand(*tf_to_shape_tuple(x))
# test
self.run(f, tf_feed_dict=feed_dict)
def test_sum_and_broadcast(self):
# placeholder
a = tf.placeholder(tf.float32, shape=[3, 4, 5, 6])
b = tf.placeholder(tf.float32, shape=[3, 4, 5])
a_sum = tf.reduce_sum(a, reduction_indices=[0, 3]) # shape (4, 5)
b_sum = tf.reduce_sum(b, reduction_indices=[0, 1]) # shape (5,)
f = a_sum + b_sum + b # (4, 5) + (5,) + (3, 4, 5) -> (3, 4, 5)
# value
feed_dict = dict()
for x in [a, b]:
feed_dict[x] = np.random.rand(*tf_to_shape_tuple(x))
# test
self.run(f, tf_feed_dict=feed_dict)
def test_tanh_sigmoid(self):
# computation
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(3, 4))
c = tf.placeholder(tf.float32, shape=(2, 4))
d = tf.sigmoid(tf.matmul(a, tf.tanh(b)))
e = tf.sigmoid(c)
f = d + e
# value
feed_dict = dict()
for x in [a, b, c]:
feed_dict[x] = np.random.rand(*tf_to_shape_tuple(x))
# test
self.run(f, tf_feed_dict=feed_dict)
def test_matmul_transpose(self):
# case 1
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(3, 4))
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
self.run(tf.matmul(a, b, transpose_a=False, transpose_b=False),
tf_feed_dict={
a: a_val,
b: b_val
})
# case 2
a = tf.placeholder(tf.float32, shape=(3, 2))
b = tf.placeholder(tf.float32, shape=(3, 4))
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
self.run(tf.matmul(a, b, transpose_a=True, transpose_b=False),
tf_feed_dict={
a: a_val,
b: b_val
})
# case 3
a = tf.placeholder(tf.float32, shape=(2, 3))
b = tf.placeholder(tf.float32, shape=(4, 3))
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
self.run(tf.matmul(a, b, transpose_a=False, transpose_b=True),
tf_feed_dict={
a: a_val,
b: b_val
})
# case 4
a = tf.placeholder(tf.float32, shape=(3, 2))
b = tf.placeholder(tf.float32, shape=(4, 3))
a_val = np.random.rand(*tf_to_shape_tuple(a))
b_val = np.random.rand(*tf_to_shape_tuple(b))
self.run(tf.matmul(a, b, transpose_a=True, transpose_b=True),
tf_feed_dict={
a: a_val,
b: b_val
})
def test_shape(self):
# shapes to test
shapes = [(1, ), (1, 2), (1, 2, 3), (1, 2, 3, 4)]
# tf placeholders
placeholders = [tf.placeholder(tf.float32, shape=s) for s in shapes]
# ranks
result_ops = [tf.shape(p) for p in placeholders]
# values
feed_dict = dict()
for x in placeholders:
feed_dict[x] = np.random.rand(*tf_to_shape_tuple(x))
# test
for op in result_ops:
self.run(op, tf_feed_dict=feed_dict)
def test_sum_mean(self):
# test cases
reduction_indices_list = [None, [], [
0,
], [0, 1], [1, 2], [0, 1, 2]]
# tf placeholder
a = tf.placeholder(tf.float32, shape=[3, 4, 5])
# value
feed_dict = {a: np.random.rand(*tf_to_shape_tuple(a))}
# test
for reduction_indices in reduction_indices_list:
f = tf.reduce_sum(a, reduction_indices=reduction_indices)
self.run(f, tf_feed_dict=feed_dict)
g = tf.reduce_mean(a, reduction_indices=reduction_indices)
self.run(g, tf_feed_dict=feed_dict)
def test_broadcast_rules(self):
# tf have un-implemented broadcasts
# for example: (2, 1, 2, 1) + (1, 2, 1, 2) is not implemented in tf
# (10, 1, 2, 1, 5) + (11, 1, 1, 5) is not implemented in tf
a = tf.placeholder(tf.float32, shape=(5, 1, 1, 4))
b = tf.placeholder(tf.float32, shape=(1, 1, 3, 1, 1))
c = tf.placeholder(tf.float32, shape=(1, 1, 4))
d = tf.placeholder(tf.float32, shape=(4, ))
f = a + b * c + d
# value
feed_dict = dict()
for x in [a, b, c, d]:
feed_dict[x] = np.random.rand(*tf_to_shape_tuple(x))
# test
self.run(f, tf_feed_dict=feed_dict)
