def c_code(self, node, name, ins, outs, sub):
# support old pickled graphs
if len(ins) == 2:
(pvals, unis) = ins
n = 1
else:
(pvals, unis, n) = ins
(z,) = outs
if self.odtype == "auto":
t = f"PyArray_TYPE({pvals})"
else:
t = Scalar(self.odtype).dtype_specs()[1]
if t.startswith("theano_complex"):
t = t.replace("theano_complex", "NPY_COMPLEX")
else:
t = t.upper()
fail = sub["fail"]
return (
"""
if (PyArray_NDIM(%(pvals)s) != 2)
{
PyErr_Format(PyExc_TypeError, "pvals ndim should be 2");
%(fail)s;
}
if (PyArray_NDIM(%(unis)s) != 1)
{
PyErr_Format(PyExc_TypeError, "unis ndim should be 2");
%(fail)s;
}
if (PyArray_DIMS(%(unis)s)[0] != (PyArray_DIMS(%(pvals)s)[0] * %(n)s))
{
PyErr_Format(PyExc_ValueError, "unis.shape[0] != pvals.shape[0] * n");
%(fail)s;
}
if ((NULL == %(z)s)
|| ((PyArray_DIMS(%(z)s))[0] != (PyArray_DIMS(%(pvals)s))[0])
|| ((PyArray_DIMS(%(z)s))[1] != (PyArray_DIMS(%(pvals)s))[1])
)
{
Py_XDECREF(%(z)s);
%(z)s = (PyArrayObject*) PyArray_EMPTY(2,
PyArray_DIMS(%(pvals)s),
%(t)s,
0);
if (!%(z)s)
{
PyErr_SetString(PyExc_MemoryError, "failed to alloc z output");
%(fail)s;
}
}
{ // NESTED SCOPE
const int nb_multi = PyArray_DIMS(%(pvals)s)[0];
const int nb_outcomes = PyArray_DIMS(%(pvals)s)[1];
const int n_samples = %(n)s;
//
// For each multinomial, loop over each possible outcome
//
for (int c = 0; c < n_samples; ++c){
for (int n = 0; n < nb_multi; ++n)
{
int waiting = 1;
double cummul = 0.;
const dtype_%(unis)s* unis_n = (dtype_%(unis)s*)PyArray_GETPTR1(%(unis)s, c*nb_multi + n);
for (int m = 0; m < nb_outcomes; ++m)
{
dtype_%(z)s* z_nm = (dtype_%(z)s*)PyArray_GETPTR2(%(z)s, n,m);
const dtype_%(pvals)s* pvals_nm = (dtype_%(pvals)s*)PyArray_GETPTR2(%(pvals)s, n,m);
cummul += *pvals_nm;
if (c == 0)
{
if (waiting && (cummul > *unis_n))
{
*z_nm = 1.;
waiting = 0;
}
else
{
// if we re-used old z pointer, we have to clear it out.
*z_nm = 0.;
}
}
else {
if (cummul > *unis_n)
{
*z_nm = *z_nm + 1.;
break;
}
}
}
}
}
} // END NESTED SCOPE
"""
% locals()
)
def c_code(self, node, name, ins, outs, sub):
(pvals, unis, n) = ins
(z,) = outs
replace = int(self.replace)
if self.odtype == "auto":
t = "NPY_INT64"
else:
t = Scalar(self.odtype).dtype_specs()[1]
if t.startswith("theano_complex"):
t = t.replace("theano_complex", "NPY_COMPLEX")
else:
t = t.upper()
fail = sub["fail"]
return (
"""
// create a copy of pvals matrix
PyArrayObject* pvals_copy = NULL;
if (PyArray_NDIM(%(pvals)s) != 2)
{
PyErr_Format(PyExc_TypeError, "pvals ndim should be 2");
%(fail)s;
}
if (PyArray_NDIM(%(unis)s) != 1)
{
PyErr_Format(PyExc_TypeError, "unis ndim should be 2");
%(fail)s;
}
if ( %(n)s > (PyArray_DIMS(%(pvals)s)[1]) )
{
PyErr_Format(PyExc_ValueError, "Cannot sample without replacement n samples bigger than the size of the distribution.");
%(fail)s;
}
if (PyArray_DIMS(%(unis)s)[0] != (PyArray_DIMS(%(pvals)s)[0] * %(n)s))
{
PyErr_Format(PyExc_ValueError, "unis.shape[0] != pvals.shape[0] * n");
%(fail)s;
}
pvals_copy = (PyArrayObject*) PyArray_EMPTY(2,
PyArray_DIMS(%(pvals)s),
PyArray_TYPE(%(pvals)s),
0);
if (!pvals_copy)
{
PyErr_SetString(PyExc_MemoryError, "failed to alloc pvals_copy");
%(fail)s;
}
PyArray_CopyInto(pvals_copy, %(pvals)s);
if ((NULL == %(z)s)
|| ((PyArray_DIMS(%(z)s))[0] != (PyArray_DIMS(%(pvals)s))[0])
|| ((PyArray_DIMS(%(z)s))[1] != %(n)s)
)
{
Py_XDECREF(%(z)s);
npy_intp dims[2];
dims[0] = PyArray_DIMS(%(pvals)s)[0];
dims[1] = %(n)s;
%(z)s = (PyArrayObject*) PyArray_EMPTY(2,
dims,
%(t)s,
-1);
if (!%(z)s)
{
PyErr_SetString(PyExc_MemoryError, "failed to alloc z output");
%(fail)s;
}
}
{ // NESTED SCOPE
const int nb_multi = PyArray_DIMS(%(pvals)s)[0];
const int nb_outcomes = PyArray_DIMS(%(pvals)s)[1];
const int n_samples = %(n)s;
//
// For each multinomial, loop over each possible outcome,
// and set selected pval to 0 after being selected
//
for (int c = 0; c < n_samples; ++c){
for (int n = 0; n < nb_multi; ++n)
{
double cummul = 0.;
const dtype_%(unis)s* unis_n = (dtype_%(unis)s*)PyArray_GETPTR1(%(unis)s, c*nb_multi + n);
dtype_%(z)s* z_nc = (dtype_%(z)s*)PyArray_GETPTR2(%(z)s, n, c);
for (int m = 0; m < nb_outcomes; ++m)
{
dtype_%(pvals)s* pvals_nm = (dtype_%(pvals)s*)PyArray_GETPTR2(pvals_copy, n, m);
cummul += *pvals_nm;
if (cummul > *unis_n)
{
*z_nc = m;
// No need to renormalize after the last samples.
if (c == (n_samples - 1))
break;
if (! %(replace)s )
{
// renormalize the nth row of pvals, reuse (cummul-*pvals_nm) to initialize the sum
dtype_%(pvals)s sum = cummul - *pvals_nm;
dtype_%(pvals)s* pvals_n = (dtype_%(pvals)s*)PyArray_GETPTR2(pvals_copy, n, m);
*pvals_nm = 0.;
for (int k = m; k < nb_outcomes; ++k)
{
sum = sum + *pvals_n;
pvals_n++;
}
pvals_n = (dtype_%(pvals)s*)PyArray_GETPTR2(pvals_copy, n, 0);
for (int k = 0; k < nb_outcomes; ++k)
{
*pvals_n = *pvals_n / sum;
pvals_n++;
}
}
break;
}
}
}
}
// delete pvals_copy
{
Py_XDECREF(pvals_copy);
}
} // END NESTED SCOPE
"""
% locals()
)
