def test_percentile_keepdims_input_broadcastable():
v = T.tensor4()
nt.assert_equal(ttt.percentile(v.dimshuffle(0, 1, 2, "x", 3),
0,
axis=[2]).broadcastable,
(False, False, True, False))
v = T.tensor4()
nt.assert_equal(ttt.percentile(v.dimshuffle(0, 1, 2, 3, "x"),
0,
axis=[2],
keepdims=True).broadcastable,
(False, False, True, False, True))
def get_output_for(self, input, deterministic=False, **kwargs):
beta = self.beta;
if not deterministic:
self_beta = theano.clone(self.beta, share_inputs=False);
input_beta = ttt.percentile(input, self.perc);
self_beta.default_update = ((1 - self.alpha) * self_beta + self.alpha * input_beta);
beta += 0 * self_beta;
# thresholding
return theano.tensor.nnet.sigmoid(self.tight*(input-beta+self.bias));
def get_output_for(self, input, deterministic=False, **kwargs):
beta = self.beta
if not deterministic:
self_beta = theano.clone(self.beta, share_inputs=False)
input_beta = ttt.percentile(input, self.perc)
self_beta.default_update = ((1 - self.alpha) * self_beta +
self.alpha * input_beta)
beta += 0 * self_beta
# thresholding
return theano.tensor.nnet.relu(input - beta, 0.0)
def test_percentile_keepdims_broadcastable():
v = T.tensor4()
x = np.array(np.random.rand(*range(4, 8)), dtype=fX)
q = 42.42
ans = x + np.percentile(x, q, axis=[2, 3], keepdims=True)
res_var = v + ttt.percentile(v,
q,
axis=[2, 3],
keepdims=True)
fn = theano.function([v], [res_var])
res = fn(x)[0]
np.testing.assert_allclose(ans, res, rtol=1e-5)
def test_percentile_axis():
v = T.tensor4()
x = np.array(np.random.rand(*range(4, 8)), dtype=fX)
q = 42.42
for axis in [None,
[2, 3],
[0, 1],
[0, 1, 2, 3],
3]:
fn = theano.function([v], [ttt.percentile(v, q, axis=axis)])
ans = np.percentile(x, q, axis=axis)
res = fn(x)[0]
np.testing.assert_allclose(ans, res, rtol=1e-5)
def test_percentile_q_constant():
# testing that percentile works when q is not a var
for dims, var in enumerate([T.scalar(),
T.vector(),
T.matrix(),
T.tensor3(),
T.tensor4()]):
for q in [0, 10, 42.3, 88, 100]:
fn = theano.function([var], [ttt.percentile(var, q)])
x = np.array(np.random.rand(*(dims * [4])), dtype=fX)
ans = np.percentile(x, q)
res = fn(x)[0]
np.testing.assert_allclose(ans, res)
def test_percentile():
q_var = T.scalar()
for dims, var in enumerate([T.scalar(),
T.vector(),
T.matrix(),
T.tensor3(),
T.tensor4()]):
fn = theano.function([var, q_var], [ttt.percentile(var, q_var)])
for q in [0, 10, 42.3, 88, 100]:
x = np.array(np.random.rand(*(dims * [4])), dtype=fX)
ans = np.percentile(x, q)
res = fn(x, q)[0]
np.testing.assert_allclose(ans, res)
