# include <cppad/cppad.hpp>
namespace { // Begin empty namespace
using CppAD::vector;
// a utility to compute the union of two sets.
void my_union(
std::set<size_t>& result ,
const std::set<size_t>& left ,
const std::set<size_t>& right )
{ std::set<size_t> temp;
std::set_union(
left.begin() ,
left.end() ,
right.begin() ,
right.end() ,
std::inserter(temp, temp.begin())
);
result.swap(temp);
}
// ----------------------------------------------------------------------
// forward mode routine called by CppAD
bool user_tan_forward(
size_t id ,
size_t order ,
size_t n ,
size_t m ,
const vector<bool>& vx ,
vector<bool>& vzy ,
const vector<float>& tx ,
vector<float>& tzy
)
{
assert( id == 0 || id == 1 );
assert( n == 1 );
assert( m == 2 );
assert( tx.size() >= (order+1) * n );
assert( tzy.size() >= (order+1) * m );
size_t n_order = order + 1;
size_t j = order;
size_t k;
// check if this is during the call to user_tan(id, ax, ay)
if( vx.size() > 0 )
{ assert( vx.size() >= n );
assert( vzy.size() >= m );
// now setvzy
vzy[0] = vx[0];
vzy[1] = vx[0];
}
if( j == 0 )
{ // z^{(0)} = tan( x^{(0)} ) or tanh( x^{(0)} )
if( id == 0 )
tzy[0] = tan( tx[0] );
else tzy[0] = tanh( tx[0] );
// y^{(0)} = z^{(0)} * z^{(0)}
tzy[n_order + 0] = tzy[0] * tzy[0];
}
else
{ float j_inv = 1.f / float(j);
if( id == 1 )
j_inv = - j_inv;
// z^{(j)} = x^{(j)} +- sum_{k=1}^j k x^{(k)} y^{(j-k)} / j
tzy[j] = tx[j];
for(k = 1; k <= j; k++)
tzy[j] += tx[k] * tzy[n_order + j-k] * k * j_inv;
// y^{(j)} = sum_{k=0}^j z^{(k)} z^{(j-k)}
tzy[n_order + j] = 0.;
for(k = 0; k <= j; k++)
tzy[n_order + j] += tzy[k] * tzy[j-k];
}
// All orders are implemented and there are no possible errors
return true;
}
// ----------------------------------------------------------------------
// reverse mode routine called by CppAD
bool user_tan_reverse(
size_t id ,
size_t order ,
size_t n ,
size_t m ,
const vector<float>& tx ,
const vector<float>& tzy ,
vector<float>& px ,
const vector<float>& pzy
)
{
assert( id == 0 || id == 1 );
assert( n == 1 );
assert( m == 2 );
assert( tx.size() >= (order+1) * n );
assert( tzy.size() >= (order+1) * m );
assert( px.size() >= (order+1) * n );
assert( pzy.size() >= (order+1) * m );
size_t n_order = order + 1;
size_t j, k;
// copy because partials w.r.t. y and z need to change
vector<float> qzy = pzy;
// initialize accumultion of reverse mode partials
for(k = 0; k < n_order; k++)
px[k] = 0.;
// eliminate positive orders
for(j = order; j > 0; j--)
{ float j_inv = 1.f / float(j);
if( id == 1 )
j_inv = - j_inv;
// H_{x^{(k)}} += delta(j-k) +- H_{z^{(j)} y^{(j-k)} * k / j
px[j] += qzy[j];
for(k = 1; k <= j; k++)
px[k] += qzy[j] * tzy[n_order + j-k] * k * j_inv;
// H_{y^{j-k)} += +- H_{z^{(j)} x^{(k)} * k / j
for(k = 1; k <= j; k++)
qzy[n_order + j-k] += qzy[j] * tx[k] * k * j_inv;
// H_{z^{(k)}} += H_{y^{(j-1)}} * z^{(j-k-1)} * 2.
for(k = 0; k < j; k++)
qzy[k] += qzy[n_order + j-1] * tzy[j-k-1] * 2.f;
}
// eliminate order zero
if( id == 0 )
px[0] += qzy[0] * (1.f + tzy[n_order + 0]);
else
px[0] += qzy[0] * (1.f - tzy[n_order + 0]);
return true;
}
// ----------------------------------------------------------------------
// forward Jacobian sparsity routine called by CppAD
bool user_tan_for_jac_sparse(
size_t id ,
size_t n ,
size_t m ,
size_t q ,
const vector< std::set<size_t> >& r ,
vector< std::set<size_t> >& s )
{
assert( n == 1 );
assert( m == 2 );
assert( id == 0 || id == 1 );
assert( r.size() >= n );
assert( s.size() >= m );
// sparsity for z and y are the same as for x
s[0] = r[0];
s[1] = r[0];
return true;
}
// ----------------------------------------------------------------------
// reverse Jacobian sparsity routine called by CppAD
bool user_tan_rev_jac_sparse(
size_t id ,
size_t n ,
size_t m ,
size_t q ,
vector< std::set<size_t> >& r ,
const vector< std::set<size_t> >& s )
{
assert( n == 1 );
assert( m == 2 );
assert( id == 0 || id == 1 );
assert( r.size() >= n );
assert( s.size() >= m );
// note that, if the users code only uses z, and not y,
// we could just set r[0] = s[0]
my_union(r[0], s[0], s[1]);
return true;
}
// ----------------------------------------------------------------------
// reverse Hessian sparsity routine called by CppAD
bool user_tan_rev_hes_sparse(
size_t id ,
size_t n ,
size_t m ,
size_t q ,
const vector< std::set<size_t> >& r ,
const vector<bool>& s ,
vector<bool>& t ,
const vector< std::set<size_t> >& u ,
vector< std::set<size_t> >& v )
{
assert( n == 1 );
assert( m == 2 );
assert( id == 0 || id == 1 );
assert( r.size() >= n );
assert( s.size() >= m );
assert( t.size() >= n );
assert( u.size() >= m );
assert( v.size() >= n );
// back propogate Jacobian sparsity. If users code only uses z,
// we could just set t[0] = s[0];
t[0] = s[0] | s[1];
// back propogate Hessian sparsity, ...
my_union(v[0], u[0], u[1]);
// convert forward Jacobian sparsity to Hessian sparsity
// because tan and tanh are nonlinear
if( t[0] )
my_union(v[0], v[0], r[0]);
return true;
}
// ---------------------------------------------------------------------
// Declare the AD<float> routine user_tan(id, ax, ay)
CPPAD_USER_ATOMIC(
user_tan ,
CPPAD_TEST_VECTOR ,
float ,
user_tan_forward ,
user_tan_reverse ,
user_tan_for_jac_sparse ,
user_tan_rev_jac_sparse ,
user_tan_rev_hes_sparse
)
} // End empty namespace
bool user_tan(void)
{ bool ok = true;
using CppAD::AD;
using CppAD::NearEqual;
float eps = 10.f * CppAD::epsilon<float>();
// domain space vector
size_t n = 1;
float x0 = 0.5;
CPPAD_TEST_VECTOR< AD<float> > ax(n);
ax[0] = x0;
// declare independent variables and start tape recording
CppAD::Independent(ax);
// range space vector
size_t m = 3;
CPPAD_TEST_VECTOR< AD<float> > af(m);
// temporary vector for user_tan computations
// (user_tan computes tan or tanh and its square)
CPPAD_TEST_VECTOR< AD<float> > az(2);
// call user tan function and store tan(x) in f[0] (ignore tan(x)^2)
size_t id = 0;
user_tan(id, ax, az);
af[0] = az[0];
// call user tanh function and store tanh(x) in f[1] (ignore tanh(x)^2)
id = 1;
user_tan(id, ax, az);
af[1] = az[0];
// put a constant in f[2] = tanh(1.) (for sparsity pattern testing)
CPPAD_TEST_VECTOR< AD<float> > one(1);
one[0] = 1.;
user_tan(id, one, az);
af[2] = az[0];
// create f: x -> f and stop tape recording
CppAD::ADFun<float> F(ax, af);
// check value
float tan = std::tan(x0);
ok &= NearEqual(af[0] , tan, eps, eps);
float tanh = std::tanh(x0);
ok &= NearEqual(af[1] , tanh, eps, eps);
// compute first partial of f w.r.t. x[0] using forward mode
CPPAD_TEST_VECTOR<float> dx(n), df(m);
dx[0] = 1.;
df = F.Forward(1, dx);
// compute derivative of tan - tanh using reverse mode
CPPAD_TEST_VECTOR<float> w(m), dw(n);
w[0] = 1.;
w[1] = 1.;
dw = F.Reverse(1, w);
// tan'(x) = 1 + tan(x) * tan(x)
// tanh'(x) = 1 - tanh(x) * tanh(x)
float tanp = 1.f + tan * tan;
float tanhp = 1.f - tanh * tanh;
ok &= NearEqual(df[0], tanp, eps, eps);
ok &= NearEqual(df[1], tanhp, eps, eps);
ok &= NearEqual(dw[0], w[0]*tanp + w[1]*tanhp, eps, eps);
// compute second partial of f w.r.t. x[0] using forward mode
CPPAD_TEST_VECTOR<float> ddx(n), ddf(m);
ddx[0] = 0.;
ddf = F.Forward(2, ddx);
// compute second derivative of tan - tanh using reverse mode
CPPAD_TEST_VECTOR<float> ddw(2);
ddw = F.Reverse(2, w);
// tan''(x) = 2 * tan(x) * tan'(x)
// tanh''(x) = - 2 * tanh(x) * tanh'(x)
// Note that second order Taylor coefficient for u half the
// corresponding second derivative.
float two = 2;
float tanpp = two * tan * tanp;
float tanhpp = - two * tanh * tanhp;
ok &= NearEqual(two * ddf[0], tanpp, eps, eps);
ok &= NearEqual(two * ddf[1], tanhpp, eps, eps);
ok &= NearEqual(ddw[0], w[0]*tanp + w[1]*tanhp , eps, eps);
ok &= NearEqual(ddw[1], w[0]*tanpp + w[1]*tanhpp, eps, eps);
// Forward mode computation of sparsity pattern for F.
size_t q = n;
// user vectorBool because m and n are small
CppAD::vectorBool r1(q), s1(m * q);
r1[0] = true; // propogate sparsity for x[0]
s1 = F.ForSparseJac(q, r1);
ok &= (s1[0] == true); // f[0] depends on x[0]
ok &= (s1[1] == true); // f[1] depends on x[0]
ok &= (s1[2] == false); // f[2] does not depend on x[0]
// Reverse mode computation of sparsity pattern for F.
size_t p = m;
CppAD::vectorBool s2(p * m), r2(p * n);
// Sparsity pattern for identity matrix
size_t i, j;
for(i = 0; i < p; i++)
{ for(j = 0; j < m; j++)
s2[i * p + j] = (i == j);
}
r2 = F.RevSparseJac(p, s2);
ok &= (r2[0] == true); // f[0] depends on x[0]
ok &= (r2[1] == true); // f[1] depends on x[0]
ok &= (r2[2] == false); // f[2] does not depend on x[0]
// Hessian sparsity for f[0]
CppAD::vectorBool s3(m), h(q * n);
s3[0] = true;
s3[1] = false;
s3[2] = false;
h = F.RevSparseHes(q, s3);
ok &= (h[0] == true); // Hessian is non-zero
// Hessian sparsity for f[2]
s3[0] = false;
s3[2] = true;
h = F.RevSparseHes(q, s3);
ok &= (h[0] == false); // Hessian is zero
// check tanh results for a large value of x
CPPAD_TEST_VECTOR<float> x(n), f(m);
x[0] = std::numeric_limits<float>::max() / two;
f = F.Forward(0, x);
tanh = 1.;
ok &= NearEqual(f[1], tanh, eps, eps);
df = F.Forward(1, dx);
tanhp = 0.;
ok &= NearEqual(df[1], tanhp, eps, eps);
// --------------------------------------------------------------------
// Free all temporary work space associated with user_atomic objects.
// (If there are future calls to user atomic functions, they will
// create new temporary work space.)
CppAD::user_atomic<float>::clear();
return ok;
}
