def entropy(self):
"""Entropy of Dirichlet distribution.
Returns:
Entropy of distribution.
"""
concentration0 = self.concentration.sum(-1)
k = self.concentration.shape[-1]
return (paddle.lgamma(self.concentration).sum(-1) -
paddle.lgamma(concentration0) -
(k - concentration0) * paddle.digamma(concentration0) -
((self.concentration - 1.0) *
paddle.digamma(self.concentration)).sum(-1))
def test_dtype_error(self):
# in static mode
with self.assertRaises(TypeError):
with static.program_guard(static.Program()):
x = static.data(name="x", shape=self._shape, dtype="int32")
out = paddle.digamma(x, name="digamma_res")
# in dynamic mode
with self.assertRaises(RuntimeError):
with fluid.dygraph.guard():
input = np.random.random(self._shape).astype("int32")
input_t = paddle.to_tensor(input)
res = paddle.digamma(input_t)
def test_in_dynamic_mode(self):
for dtype in self.dtypes:
input = np.random.random(self._shape).astype(dtype)
sc_res = psi(input)
for place in self.places:
# it is more convenient to use `guard` than `enable/disable_**` here
with fluid.dygraph.guard(place):
input_t = paddle.to_tensor(input)
res = paddle.digamma(input_t).numpy()
self.assertEqual(np.allclose(res, sc_res, rtol=1e-05),
True)
def test_in_static_mode(self):
def init_input_output(dtype):
input = np.random.random(self._shape).astype(dtype)
return {'x': input}, psi(input)
for dtype in self.dtypes:
input_dict, sc_res = init_input_output(dtype)
for place in self.places:
with static.program_guard(static.Program()):
x = static.data(name="x", shape=self._shape, dtype=dtype)
out = paddle.digamma(x)
exe = static.Executor(place)
out_value = exe.run(feed=input_dict, fetch_list=[out.name])
self.assertEqual(
np.allclose(out_value[0], sc_res, rtol=1e-5), True)
def test_name_argument(self):
with static.program_guard(static.Program()):
x = static.data(name="x", shape=self._shape, dtype=self.dtypes[0])
out = paddle.digamma(x, name="digamma_res")
self.assertTrue("digamma_res" in out.name)
