def test_norm_logpdf():
x = npr.randn()
l = npr.randn()
scale = npr.rand()**2 + 1.1
fun = autograd.scipy.stats.norm.logpdf
d_fun = grad(fun)
check_grads(fun, x, l, scale)
check_grads(d_fun, x, l, scale)
def test_logsumexp6():
x = npr.randn(1, 5)
def f(a):
return autograd.scipy.misc.logsumexp(a, axis=1, keepdims=True)
check_grads(f, x)
check_grads(lambda a: to_scalar(grad(f)(a)), x)
def test_norm_logpdf():
x = npr.randn()
l = npr.randn()
scale=npr.rand()**2 + 1.1
fun = autograd.scipy.stats.norm.logpdf
d_fun = grad(fun)
check_grads(fun, x, l, scale)
check_grads(d_fun, x, l, scale)
def test_logsumexp6():
x = npr.randn(1, 5)
def f(a):
return autograd.scipy.misc.logsumexp(a, axis=1, keepdims=True)
check_grads(f, x)
check_grads(lambda a: to_scalar(grad(f)(a)), x)
def test_yn():
x = npr.randn()**2 + 0.3
fun = lambda x: to_scalar(autograd.scipy.special.yn(2, x))
d_fun = grad(fun)
check_grads(fun, x)
check_grads(d_fun, x)
def test_polygamma():
x = npr.randn()
fun = lambda x: to_scalar(autograd.scipy.special.polygamma(0, x))
d_fun = grad(fun)
check_grads(fun, x)
check_grads(d_fun, x)
def test_yn():
x = npr.randn()**2 + 0.3
fun = lambda x: to_scalar(autograd.scipy.special.yn(2, x))
d_fun = grad(fun)
check_grads(fun, x)
check_grads(d_fun, x)
def test_polygamma():
x = npr.randn()
fun = lambda x: to_scalar(autograd.scipy.special.polygamma(0, x))
d_fun = grad(fun)
check_grads(fun, x)
check_grads(d_fun, x)
