float simple_liner_regression(np::ndarray a, np::ndarray b, np::ndarray c) {
int nd1 = a.get_nd();
int nd2 = b.get_nd();
if (nd1 != 1 || nd2 != 1)
throw std::runtime_error("a and b must be 1-dimensional");
if ( (a.get_dtype() != np::dtype::get_builtin<double>()) ||
(b.get_dtype() != np::dtype::get_builtin<double>()) )
throw std::runtime_error("a and b must be float64 array");
size_t N = a.shape(0);
if ( N != b.shape(0) )
throw std::runtime_error(" a and b must be same size");
double *p = reinterpret_cast<double *>(a.get_data());
std::vector<float> x;
for(int i=0;i<N;i++) x.push_back(*p++);
double *q = reinterpret_cast<double *>(b.get_data());
std::vector<float> y;
for(int i=0;i<N;i++) y.push_back(*q++);
// 回帰系数の計算
float a1 = calc_covariance(x,y) / calc_variance(x);
float a0 = calc_mean(y) - a1 * calc_mean(x);
double *r = reinterpret_cast<double *>(c.get_data());
*r = a0; r++;
*r = a1;
}
int relax_precision(const np::ndarray& predict,
const np::ndarray& label, const int& relax) {
const int h_lim = predict.shape(1);
const int w_lim = predict.shape(0);
const int32_t *predict_data =
reinterpret_cast<int32_t *>(predict.get_data());
const int32_t *label_data =
reinterpret_cast<int32_t *>(label.get_data());
int true_positive = 0;
for (int y = 0; y < h_lim; ++y) {
for (int x = 0; x < w_lim; ++x) {
const int32_t pred_val = predict_data[y * w_lim + x];
if (pred_val == 1) {
const int st_y = y - relax >= 0 ? y - relax : 0;
const int en_y = y + relax < h_lim ? y + relax : h_lim - 1;
const int st_x = x - relax >= 0 ? x - relax : 0;
const int en_x = x + relax < w_lim ? x + relax : w_lim - 1;
int sum = 0;
for (int yy = st_y; yy <= en_y; ++yy) {
for (int xx = st_x; xx <= en_x; ++xx) {
sum += label_data[yy * w_lim + xx];
}
}
if (sum > 0) true_positive++;
}
}
}
return true_positive;
}
//Second Constructor
//Initializes a CNet instance based on
// the passed options struct (tuple(list(dict)), see CNet_getopts)
std::shared_ptr<network> CNet_loadopts( bp::tuple const & opts,
std::string const net_config_file,
np::ndarray const & outsz_a,
std::size_t const tc,
bool const is_optimize = true,
std::uint8_t const phs = 0,
bool const force_fft = false )
{
bp::list node_opts_list = bp::extract<bp::list>( opts[0] );
bp::list edge_opts_list = bp::extract<bp::list>( opts[1] );
//See pyznn_utils.hpp
std::vector<options> node_opts = pyopt_to_znnopt(node_opts_list);
std::vector<options> edge_opts = pyopt_to_znnopt(edge_opts_list);
vec3i out_sz( reinterpret_cast<std::int64_t*>(outsz_a.get_data())[0],
reinterpret_cast<std::int64_t*>(outsz_a.get_data())[1],
reinterpret_cast<std::int64_t*>(outsz_a.get_data())[2]
);
// force fft or optimize
if ( force_fft )
{
network::force_fft(edge_opts);
}
else
{
if ( is_optimize )
{
phase _phs = static_cast<phase>(phs);
if ( _phs == phase::TRAIN )
{
network::optimize(node_opts, edge_opts, out_sz, tc, 10);
}
else if ( _phs == phase::TEST )
{
network::optimize_forward(node_opts, edge_opts, out_sz, tc, 2);
}
else
{
std::string str = boost::lexical_cast<std::string>(phs);
throw std::logic_error(HERE() + "unknown phase: " + str);
}
}
}
std::shared_ptr<network> net(
new network( node_opts,edge_opts,out_sz,tc,static_cast<phase>(phs) ));
return net;
}
//===========================================================================
//IO FUNCTIONS
//First constructor - generates a random network
std::shared_ptr< network > CNet_Init(
std::string const net_config_file,
np::ndarray const & outsz_a,
std::size_t tc = 0, // thread number
bool const is_optimize = true,
std::uint8_t const phs = 0, // 0:TRAIN, 1:TEST
bool const force_fft = false)
{
std::vector<options> nodes;
std::vector<options> edges;
std::cout<< "parse_net file: "<<net_config_file<<std::endl;
parse_net_file(nodes, edges, net_config_file);
vec3i out_sz( reinterpret_cast<std::int64_t*>(outsz_a.get_data())[0],
reinterpret_cast<std::int64_t*>(outsz_a.get_data())[1],
reinterpret_cast<std::int64_t*>(outsz_a.get_data())[2]
);
if ( tc == 0 )
tc = std::thread::hardware_concurrency();
// force fft or optimize
if ( force_fft )
{
network::force_fft(edges);
}
else
{
if ( is_optimize )
{
phase _phs = static_cast<phase>(phs);
if ( _phs == phase::TRAIN )
{
network::optimize(nodes, edges, out_sz, tc, 10);
}
else if ( _phs == phase::TEST )
{
network::optimize_forward(nodes, edges, out_sz, tc, 2);
}
else
{
std::string str = boost::lexical_cast<std::string>(phs);
throw std::logic_error(HERE() + "unknown phase: " + str);
}
}
}
std::cout<< "construct the network class using the edges and nodes..." <<std::endl;
// construct the network class
std::shared_ptr<network> net(
new network(nodes,edges,out_sz,tc,static_cast<phase>(phs)));
return net;
}
/**
* Here, we'll take the alternate approach of using the strides to access the array's memory directly.
* This can be much faster for large arrays.
*/
static void copy_ndarray_to_mv2(bn::ndarray const & array, matrix2 & mat) {
// Unfortunately, get_strides() can't be inlined, so it's best to call it once up-front.
Py_intptr_t const * strides = array.get_strides();
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < 2; ++j) {
mat(i, j) = *reinterpret_cast<double const *>(array.get_data() + i * strides[0] + j * strides[1]);
}
}
}
// Here's a simple wrapper function for fill1. It requires that the passed
// NumPy array be exactly what we're looking for - no conversion from nested
// sequences or arrays with other data types, because we want to modify it
// in-place.
void wrap_fill1(np::ndarray const & array) {
if (array.get_dtype() != np::dtype::get_builtin<double>()) {
PyErr_SetString(PyExc_TypeError, "Incorrect array data type");
p::throw_error_already_set();
}
if (array.get_nd() != 2) {
PyErr_SetString(PyExc_TypeError, "Incorrect number of dimensions");
p::throw_error_already_set();
}
fill1(reinterpret_cast<double*>(array.get_data()),
array.shape(0), array.shape(1),
array.strides(0) / sizeof(double), array.strides(1) / sizeof(double));
}
NumPyArrayData<T>(const np::ndarray &arr){
np::dtype dtype = arr.get_dtype();
np::dtype dtype_expected = np::dtype::get_builtin<T>();
if(dtype != dtype_expected) {
std::stringstream ss;
ss << "NumPyArrayData: Unexpected data type (" << bp::extract<const char*>(dtype.attr("__str__")()) << ") received";
ss << "Expected " << bp::extract<const char*>(dtype_expected.attr("__str__")());
throw std::runtime_error(ss.str().c_str());
}
data_ = arr.get_data();
strides_ = arr.get_strides();
}
float u_variance(np::ndarray a) {
int nd = a.get_nd();
if (nd != 1)
throw std::runtime_error("a must be 1-dimensional");
if (a.get_dtype() != np::dtype::get_builtin<double>())
throw std::runtime_error("a must be float64 array");
size_t N = a.shape(0);
double *p = reinterpret_cast<double *>(a.get_data());
std::vector<float> x;
for(int i=0;i<N;i++) x.push_back(*p++);
return calc_u_variance(x);
}
float covariance(np::ndarray a, np::ndarray b) {
int nd1 = a.get_nd();
int nd2 = b.get_nd();
if (nd1 != 1 || nd2 != 1)
throw std::runtime_error("a and b must be 1-dimensional");
if ( (a.get_dtype() != np::dtype::get_builtin<double>()) ||
(b.get_dtype() != np::dtype::get_builtin<double>()) )
throw std::runtime_error("a and b must be float64 array");
size_t N = a.shape(0);
if ( N != b.shape(0) )
throw std::runtime_error(" a and b must be same size");
double *p = reinterpret_cast<double *>(a.get_data());
std::vector<float> x;
for(int i=0;i<N;i++) x.push_back(*p++);
double *q = reinterpret_cast<double *>(b.get_data());
std::vector<float> y;
for(int i=0;i<N;i++) y.push_back(*q++);
return calc_covariance(x,y);
}
// Here's the wrapper for fill2; it's a little more complicated because we need
// to check the flags and create the array of pointers.
void wrap_fill2(np::ndarray const & array) {
if (array.get_dtype() != np::dtype::get_builtin<double>()) {
PyErr_SetString(PyExc_TypeError, "Incorrect array data type");
p::throw_error_already_set();
}
if (array.get_nd() != 2) {
PyErr_SetString(PyExc_TypeError, "Incorrect number of dimensions");
p::throw_error_already_set();
}
if (!(array.get_flags() & np::ndarray::C_CONTIGUOUS)) {
PyErr_SetString(PyExc_TypeError, "Array must be row-major contiguous");
p::throw_error_already_set();
}
double * iter = reinterpret_cast<double*>(array.get_data());
int rows = array.shape(0);
int cols = array.shape(1);
boost::scoped_array<double*> ptrs(new double*[rows]);
for (int i = 0; i < rows; ++i, iter += cols) {
ptrs[i] = iter;
}
fill2(ptrs.get(), array.shape(0), array.shape(1));
}