pylibvw.cc

#include "../vowpalwabbit/vw.h"
#include "../vowpalwabbit/multiclass.h"
#include "../vowpalwabbit/cost_sensitive.h"
#include "../vowpalwabbit/cb.h"
#include "../vowpalwabbit/search.h"
#include "../vowpalwabbit/search_hooktask.h"
#include "../vowpalwabbit/parse_example.h"
#include "../vowpalwabbit/gd.h"

#include <boost/make_shared.hpp>
#include <boost/python.hpp>
#include <boost/python/suite/indexing/vector_indexing_suite.hpp>

using namespace std;
namespace py=boost::python;

typedef boost::shared_ptr<vw> vw_ptr;
typedef boost::shared_ptr<example> example_ptr;
typedef boost::shared_ptr<Search::search> search_ptr;
typedef boost::shared_ptr<Search::predictor> predictor_ptr;

const size_t lDEFAULT = 0;
const size_t lBINARY = 1;
const size_t lMULTICLASS = 2;
const size_t lCOST_SENSITIVE = 3;
const size_t lCONTEXTUAL_BANDIT = 4;
const size_t lMAX = 5;


void dont_delete_me(void*arg) { }

vw_ptr my_initialize(string args)
{ vw*foo = VW::initialize(args);
  return boost::shared_ptr<vw>(foo, dont_delete_me);
}

void my_finish(vw_ptr all)
{ VW::finish(*all, false);  // don't delete all because python will do that for us!
}

search_ptr get_search_ptr(vw_ptr all)
{ return boost::shared_ptr<Search::search>((Search::search*)(all->searchstr), dont_delete_me);
}

void my_audit_example(vw_ptr all, example_ptr ec) { GD::print_audit_features(*all, *ec); }

predictor_ptr get_predictor(search_ptr sch, ptag my_tag)
{ Search::predictor* P = new Search::predictor(*sch, my_tag);
  return boost::shared_ptr<Search::predictor>(P);
}

label_parser* get_label_parser(vw*all, size_t labelType)
{ switch (labelType)
  { case lDEFAULT:           return all ? &all->p->lp : NULL;
    case lBINARY:            return &simple_label;
    case lMULTICLASS:        return &MULTICLASS::mc_label;
    case lCOST_SENSITIVE:    return &COST_SENSITIVE::cs_label;
    case lCONTEXTUAL_BANDIT: return &CB::cb_label;
    default: cerr << "get_label_parser called on invalid label type" << endl; throw exception();
  }
}

void my_delete_example(void*voidec)
{ example* ec = (example*) voidec;
  size_t labelType = (ec->tag.size() == 0) ? lDEFAULT : ec->tag[0];
  label_parser* lp = get_label_parser(NULL, labelType);
  VW::dealloc_example(lp ? lp->delete_label : NULL, *ec);
  free(ec);
}


example* my_empty_example0(vw_ptr vw, size_t labelType)
{ label_parser* lp = get_label_parser(&*vw, labelType);
  example* ec = VW::alloc_examples(lp->label_size, 1);
  lp->default_label(&ec->l);
  if (labelType == lCOST_SENSITIVE)
  { COST_SENSITIVE::wclass zero = { 0., 1, 0., 0. };
    ec->l.cs.costs.push_back(zero);
  }
  ec->tag.erase();
  if (labelType != lDEFAULT)
    ec->tag.push_back((char)labelType);  // hide the label type in the tag
  return ec;
}

example_ptr my_empty_example(vw_ptr vw, size_t labelType)
{ example* ec = my_empty_example0(vw, labelType);
  return boost::shared_ptr<example>(ec, my_delete_example);
}

example_ptr my_read_example(vw_ptr all, size_t labelType, char*str)
{ example*ec = my_empty_example0(all, labelType);
  VW::read_line(*all, ec, str);
  VW::parse_atomic_example(*all, ec, false);
  VW::setup_example(*all, ec);
  ec->example_counter = labelType;
  ec->tag.erase();
  if (labelType != lDEFAULT)
    ec->tag.push_back((char)labelType);  // hide the label type in the tag
  return boost::shared_ptr<example>(ec, my_delete_example);
}

void my_finish_example(vw_ptr all, example_ptr ec)
{ // TODO
}

void my_learn(vw_ptr all, example_ptr ec)
{ if (ec->test_only)
  { all->l->predict(*ec);
  }
  else
  { all->learn(ec.get());
  }
}

float my_learn_string(vw_ptr all, char*str)
{ example*ec = VW::read_example(*all, str);
  all->learn(ec);
  float pp = ec->partial_prediction;
  VW::finish_example(*all, ec);
  return pp;
}

float my_predict(vw_ptr all, example_ptr ec)
{ bool old_test_only = ec->test_only;
  ec->test_only = true;
  all->l->learn(*ec);
  ec->test_only = old_test_only;
  return ec->partial_prediction;
}

float my_predict_string(vw_ptr all, char*str)
{ example*ec = VW::read_example(*all, str);
  ec->test_only = true;
  all->learn(ec);
  float pp = ec->partial_prediction;
  VW::finish_example(*all, ec);
  return pp;
}

string varray_char_to_string(v_array<char> &a)
{ string ret = "";
  for (auto c : a)
    ret += c;
  return ret;
}

string my_get_tag(example_ptr ec)
{ return varray_char_to_string(ec->tag);
}

uint32_t ex_num_namespaces(example_ptr ec)
{ return ec->indices.size();
}

unsigned char ex_namespace(example_ptr ec, uint32_t ns)
{ return ec->indices[ns];
}

uint32_t ex_num_features(example_ptr ec, unsigned char ns)
{ return ec->feature_space[ns].size();
}

uint32_t ex_feature(example_ptr ec, unsigned char ns, uint32_t i)
{ return ec->feature_space[ns].indicies[i];
}

float ex_feature_weight(example_ptr ec, unsigned char ns, uint32_t i)
{ return ec->feature_space[ns].values[i];
}

float ex_sum_feat_sq(example_ptr ec, unsigned char ns)
{ return ec->feature_space[ns].sum_feat_sq;
}

void ex_push_feature(example_ptr ec, unsigned char ns, uint32_t fid, float v)
{ // warning: assumes namespace exists!
  ec->feature_space[ns].push_back(v,fid);
  ec->num_features++;
  ec->total_sum_feat_sq += v * v;
}

void ex_push_feature_list(example_ptr ec, vw_ptr vw, unsigned char ns, py::list& a)
{ // warning: assumes namespace exists!
  char ns_str[2] = { (char)ns, 0 };
  uint32_t ns_hash = VW::hash_space(*vw, ns_str);
  size_t count = 0; float sum_sq = 0.;
  for (size_t i=0; i<len(a); i++)
  { feature f = { 1., 0 };
    py::object ai = a[i];
    py::extract<py::tuple> get_tup(ai);
    if (get_tup.check())
    { py::tuple fv = get_tup();
      if (len(fv) != 2) { cerr << "warning: malformed feature in list" << endl; continue; } // TODO str(ai)
      py::extract<float> get_val(fv[1]);
      if (get_val.check())
        f.x = get_val();
      else { cerr << "warning: malformed feature in list" << endl; continue; }
      ai = fv[0];
    }

    if (f.x != 0.)
    { bool got = false;
      py::extract<string> get_str(ai);
      if (get_str.check())
      { f.weight_index = VW::hash_feature(*vw, get_str(), ns_hash);
        got = true;
      }
      else
      { py::extract<uint32_t> get_int(ai);
        if (get_int.check()) { f.weight_index = get_int(); got = true; }
        else { cerr << "warning: malformed feature in list" << endl; continue; }
      }
      if (got)
	{ ec->feature_space[ns].push_back(f.x, f.weight_index);
        count++;
	sum_sq += f.x*f.x;
      }
    }
  }
  ec->num_features += count;
  ec->total_sum_feat_sq += sum_sq;
}

void ex_push_namespace(example_ptr ec, unsigned char ns)
{ ec->indices.push_back(ns);
}

void ex_ensure_namespace_exists(example_ptr ec, unsigned char ns)
{ for (auto nss : ec->indices)
    if (ns == nss) return;
  ex_push_namespace(ec, ns);
}

void ex_push_dictionary(example_ptr ec, vw_ptr vw, py::dict& dict)
{ py::object objectKey, objectVal;
  const py::object objectKeys = dict.iterkeys();
  const py::object objectVals = dict.itervalues();
  unsigned long ulCount = boost::python::extract<unsigned long>(dict.attr("__len__")());
  for (size_t u=0; u<ulCount; u++)
  { objectKey = objectKeys.attr( "next" )();
    objectVal = objectVals.attr( "next" )();

    char chCheckKey = objectKey.ptr()->ob_type->tp_name[0];
    if (chCheckKey != 's') continue;
    chCheckKey = objectVal.ptr()->ob_type->tp_name[0];
    if (chCheckKey != 'l') continue;

    py::extract<string> ns_e(objectKey);
    if (ns_e().length() < 1) continue;
    py::extract<py::list> list_e(objectVal);
    py::list list = list_e();
    char ns = ns_e()[0];
    ex_ensure_namespace_exists(ec, ns);
    ex_push_feature_list(ec, vw, ns, list);
  }
}

bool ex_pop_feature(example_ptr ec, unsigned char ns)
{ if (ec->feature_space[ns].size() == 0) return false;
  float val = ec->feature_space[ns].values.pop();
  if (ec->feature_space[ns].indicies.size()> 0)
    ec->feature_space[ns].indicies.pop();
  if (ec->feature_space[ns].space_names.size()> 0)
    ec->feature_space[ns].space_names.pop();
  ec->num_features--;
  ec->feature_space[ns].sum_feat_sq -= val * val;
  ec->total_sum_feat_sq -= val * val;
  return true;
}

void ex_erase_namespace(example_ptr ec, unsigned char ns)
{ ec->num_features -= ec->feature_space[ns].size();
  ec->total_sum_feat_sq -= ec->feature_space[ns].sum_feat_sq;
  ec->feature_space[ns].sum_feat_sq = 0.;
  ec->feature_space[ns].erase();
}

bool ex_pop_namespace(example_ptr ec)
{ if (ec->indices.size() == 0) return false;
  unsigned char ns = ec->indices.pop();
  ex_erase_namespace(ec, ns);
  return true;
}

void my_setup_example(vw_ptr vw, example_ptr ec)
{ VW::setup_example(*vw, ec.get());
}

void unsetup_example(vw_ptr vwP, example_ptr ae)
{ vw&all = *vwP;
  ae->partial_prediction = 0.;
  ae->num_features = 0;
  ae->total_sum_feat_sq = 0;
  ae->loss = 0.;

  if (all.ignore_some)
  { cerr << "error: cannot unsetup example when some namespaces are ignored!" << endl;
    throw exception();
  }

  if(all.ngram_strings.size() > 0)
  { cerr << "error: cannot unsetup example when ngrams are in use!" << endl;
    throw exception();
  }

  if (all.add_constant)
    { ae->feature_space[constant_namespace].erase();
    int hit_constant = -1;
    size_t N = ae->indices.size();
    for (size_t i=0; i<N; i++)
    { size_t j = N - 1 - i;
      if (ae->indices[j] == constant_namespace)
      { if (hit_constant >= 0) { cerr << "error: hit constant namespace twice!" << endl; throw exception(); }
        hit_constant = j;
        break;
      }
    }
    if (hit_constant >= 0)
    { for (size_t i=hit_constant; i<N-1; i++)
        ae->indices[i] = ae->indices[i+1];
      ae->indices.pop();
    }
  }

  uint32_t multiplier = all.wpp << all.reg.stride_shift;
  if(multiplier != 1)   //make room for per-feature information.
    for (auto ns : ae->indices)
      for (auto& idx : ae->feature_space[ns].indicies)
        idx /= multiplier;
}


void ex_set_label_string(example_ptr ec, vw_ptr vw, string label, size_t labelType)
{ // SPEEDUP: if it's already set properly, don't modify
  label_parser& old_lp = vw->p->lp;
  vw->p->lp = *get_label_parser(&*vw, labelType);
  VW::parse_example_label(*vw, *ec, label);
  vw->p->lp = old_lp;
}

float ex_get_simplelabel_label(example_ptr ec) { return ec->l.simple.label; }
float ex_get_simplelabel_weight(example_ptr ec) { return ec->l.simple.weight; }
float ex_get_simplelabel_initial(example_ptr ec) { return ec->l.simple.initial; }
float ex_get_simplelabel_prediction(example_ptr ec) { return ec->pred.scalar; }

uint32_t ex_get_multiclass_label(example_ptr ec) { return ec->l.multi.label; }
float ex_get_multiclass_weight(example_ptr ec) { return ec->l.multi.weight; }
uint32_t ex_get_multiclass_prediction(example_ptr ec) { return ec->pred.multiclass; }

uint32_t ex_get_costsensitive_prediction(example_ptr ec) { return ec->pred.multiclass; }
uint32_t ex_get_costsensitive_num_costs(example_ptr ec) { return ec->l.cs.costs.size(); }
float ex_get_costsensitive_cost(example_ptr ec, uint32_t i) { return ec->l.cs.costs[i].x; }
uint32_t ex_get_costsensitive_class(example_ptr ec, uint32_t i) { return ec->l.cs.costs[i].class_index; }
float ex_get_costsensitive_partial_prediction(example_ptr ec, uint32_t i) { return ec->l.cs.costs[i].partial_prediction; }
float ex_get_costsensitive_wap_value(example_ptr ec, uint32_t i) { return ec->l.cs.costs[i].wap_value; }

uint32_t ex_get_cbandits_prediction(example_ptr ec) { return ec->pred.multiclass; }
uint32_t ex_get_cbandits_num_costs(example_ptr ec) { return ec->l.cb.costs.size(); }
float ex_get_cbandits_cost(example_ptr ec, uint32_t i) { return ec->l.cb.costs[i].cost; }
uint32_t ex_get_cbandits_class(example_ptr ec, uint32_t i) { return ec->l.cb.costs[i].action; }
float ex_get_cbandits_probability(example_ptr ec, uint32_t i) { return ec->l.cb.costs[i].probability; }
float ex_get_cbandits_partial_prediction(example_ptr ec, uint32_t i) { return ec->l.cb.costs[i].partial_prediction; }

size_t   get_example_counter(example_ptr ec) { return ec->example_counter; }
uint32_t get_ft_offset(example_ptr ec) { return ec->ft_offset; }
size_t   get_num_features(example_ptr ec) { return ec->num_features; }
float    get_partial_prediction(example_ptr ec) { return ec->partial_prediction; }
float    get_updated_prediction(example_ptr ec) { return ec->updated_prediction; }
float    get_loss(example_ptr ec) { return ec->loss; }
float    get_example_t(example_ptr ec) { return ec->example_t; }
float    get_total_sum_feat_sq(example_ptr ec) { return ec->total_sum_feat_sq; }

double get_sum_loss(vw_ptr vw) { return vw->sd->sum_loss; }
double get_weighted_examples(vw_ptr vw) { return vw->sd->weighted_examples; }

bool search_should_output(search_ptr sch) { return sch->output().good(); }
void search_output(search_ptr sch, string s) { sch->output() << s; }

/*
uint32_t search_predict_one_all(search_ptr sch, example_ptr ec, uint32_t one_ystar) {
  return sch->predict(ec.get(), one_ystar, NULL);
}

uint32_t search_predict_one_some(search_ptr sch, example_ptr ec, uint32_t one_ystar, vector<uint32_t>& yallowed) {
  v_array<uint32_t> yallowed_va;
  yallowed_va.begin       = yallowed.data();
  yallowed_va.end         = yallowed_va.begin + yallowed.size();
  yallowed_va.end_array   = yallowed_va.end;
  yallowed_va.erase_count = 0;
  return sch->predict(ec.get(), one_ystar, &yallowed_va);
}

uint32_t search_predict_many_all(search_ptr sch, example_ptr ec, vector<uint32_t>& ystar) {
  v_array<uint32_t> ystar_va;
  ystar_va.begin       = ystar.data();
  ystar_va.end         = ystar_va.begin + ystar.size();
  ystar_va.end_array   = ystar_va.end;
  ystar_va.erase_count = 0;
  return sch->predict(ec.get(), &ystar_va, NULL);
}

uint32_t search_predict_many_some(search_ptr sch, example_ptr ec, vector<uint32_t>& ystar, vector<uint32_t>& yallowed) {
  v_array<uint32_t> ystar_va;
  ystar_va.begin       = ystar.data();
  ystar_va.end         = ystar_va.begin + ystar.size();
  ystar_va.end_array   = ystar_va.end;
  ystar_va.erase_count = 0;
  v_array<uint32_t> yallowed_va;
  yallowed_va.begin       = yallowed.data();
  yallowed_va.end         = yallowed_va.begin + yallowed.size();
  yallowed_va.end_array   = yallowed_va.end;
  yallowed_va.erase_count = 0;
  return sch->predict(ec.get(), &ystar_va, &yallowed_va);
}
*/

void verify_search_set_properly(search_ptr sch)
{ if (sch->task_name == NULL)
  { cerr << "set_structured_predict_hook: search task not initialized properly" << endl;
    throw exception();
  }
  if (strcmp(sch->task_name, "hook") != 0)
  { cerr << "set_structured_predict_hook: trying to set hook when search task is not 'hook'!" << endl;
    throw exception();
  }
}

uint32_t search_get_num_actions(search_ptr sch)
{ verify_search_set_properly(sch);
  HookTask::task_data* d = sch->get_task_data<HookTask::task_data>();
  return d->num_actions;
}

void search_run_fn(Search::search&sch)
{ try
  { HookTask::task_data* d = sch.get_task_data<HookTask::task_data>();
    py::object run = *(py::object*)d->run_object;
    run.attr("__call__")();
  }
  catch(...)
  { PyErr_Print();
    PyErr_Clear();
    throw exception();
  }
}

void search_setup_fn(Search::search&sch)
{ try
  { HookTask::task_data* d = sch.get_task_data<HookTask::task_data>();
    py::object run = *(py::object*)d->setup_object;
    run.attr("__call__")();
  }
  catch(...)
  { PyErr_Print();
    PyErr_Clear();
    throw exception();
  }
}

void search_takedown_fn(Search::search&sch)
{ try
  { HookTask::task_data* d = sch.get_task_data<HookTask::task_data>();
    py::object run = *(py::object*)d->takedown_object;
    run.attr("__call__")();
  }
  catch(...)
  { PyErr_Print();
    PyErr_Clear();
    throw exception();
  }
}

void py_delete_run_object(void* pyobj)
{ py::object* o = (py::object*)pyobj;
  delete o;
}

void set_force_oracle(search_ptr sch, bool useOracle)
{ verify_search_set_properly(sch);
  sch->set_force_oracle(useOracle);
}

void set_structured_predict_hook(search_ptr sch, py::object run_object, py::object setup_object, py::object takedown_object)
{ verify_search_set_properly(sch);
  HookTask::task_data* d = sch->get_task_data<HookTask::task_data>();
  d->run_f = &search_run_fn;
  delete (py::object*)d->run_object; d->run_object = NULL;
  delete (py::object*)d->setup_object; d->setup_object = NULL;
  delete (py::object*)d->takedown_object; d->takedown_object = NULL;
  sch->set_force_oracle(false);
  d->run_object = new py::object(run_object);
  if (setup_object.ptr() != Py_None)
  { d->setup_object = new py::object(setup_object);
    d->run_setup_f = &search_setup_fn;
  }
  if (takedown_object.ptr() != Py_None)
  { d->takedown_object = new py::object(takedown_object);
    d->run_takedown_f = &search_takedown_fn;
  }
  d->delete_run_object = &py_delete_run_object;
}

void my_set_test_only(example_ptr ec, bool val) { ec->test_only = val; }

bool po_exists(search_ptr sch, string arg)
{ HookTask::task_data* d = sch->get_task_data<HookTask::task_data>();
  return (*d->var_map).count(arg) > 0;
}

string po_get_string(search_ptr sch, string arg)
{ HookTask::task_data* d = sch->get_task_data<HookTask::task_data>();
  return (*d->var_map)[arg].as<string>();
}

int32_t po_get_int(search_ptr sch, string arg)
{ HookTask::task_data* d = sch->get_task_data<HookTask::task_data>();
  try { return (*d->var_map)[arg].as<int>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<size_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<uint32_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<uint64_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<uint16_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<int32_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<int64_t>(); }
  catch (...) {}
  try { return (*d->var_map)[arg].as<int16_t>(); }
  catch (...) {}
  // we know this'll fail but do it anyway to get the exception
  return (*d->var_map)[arg].as<int>();
}

PyObject* po_get(search_ptr sch, string arg)
{ try
  { return py::incref(py::object(po_get_string(sch, arg)).ptr());
  }
  catch (...) {}
  try
  { return py::incref(py::object(po_get_int(sch, arg)).ptr());
  }
  catch (...) {}
  // return None
  return py::incref(py::object().ptr());
}

void my_set_input(predictor_ptr P, example_ptr ec) { P->set_input(*ec); }
void my_set_input_at(predictor_ptr P, size_t posn, example_ptr ec) { P->set_input_at(posn, *ec); }

void my_add_oracle(predictor_ptr P, action a) { P->add_oracle(a); }
void my_add_oracles(predictor_ptr P, py::list& a) { for (size_t i=0; i<len(a); i++) P->add_oracle(py::extract<action>(a[i])); }
void my_add_allowed(predictor_ptr P, action a) { P->add_allowed(a); }
void my_add_alloweds(predictor_ptr P, py::list& a) { for (size_t i=0; i<len(a); i++) P->add_allowed(py::extract<action>(a[i])); }
void my_add_condition(predictor_ptr P, ptag t, char c) { P->add_condition(t, c); }
void my_add_condition_range(predictor_ptr P, ptag hi, ptag count, char name0) { P->add_condition_range(hi, count, name0); }
void my_set_oracle(predictor_ptr P, action a) { P->set_oracle(a); }
void my_set_oracles(predictor_ptr P, py::list& a) { if (len(a) > 0) P->set_oracle(py::extract<action>(a[0])); else P->erase_oracles(); for (size_t i=1; i<len(a); i++) P->add_oracle(py::extract<action>(a[i])); }
void my_set_allowed(predictor_ptr P, action a) { P->set_allowed(a); }
void my_set_alloweds(predictor_ptr P, py::list& a) { if (len(a) > 0) P->set_allowed(py::extract<action>(a[0])); else P->erase_alloweds(); for (size_t i=1; i<len(a); i++) P->add_allowed(py::extract<action>(a[i])); }
void my_set_condition(predictor_ptr P, ptag t, char c) { P->set_condition(t, c); }
void my_set_condition_range(predictor_ptr P, ptag hi, ptag count, char name0) { P->set_condition_range(hi, count, name0); }
void my_set_learner_id(predictor_ptr P, size_t id) { P->set_learner_id(id); }
void my_set_tag(predictor_ptr P, ptag t) { P->set_tag(t); }


BOOST_PYTHON_MODULE(pylibvw)
{ // This will enable user-defined docstrings and python signatures,
  // while disabling the C++ signatures
  py::docstring_options local_docstring_options(true, true, false);

  // define the vw class
  py::class_<vw, vw_ptr>("vw", "the basic VW object that holds with weight vector, parser, etc.", py::no_init)
  .def("__init__", py::make_constructor(my_initialize))
  //      .def("__del__", &my_finish, "deconstruct the VW object by calling finish")
  .def("finish", &my_finish, "stop VW by calling finish (and, eg, write weights to disk)")
  .def("learn", &my_learn, "given a pyvw example, learn (and predict) on that example")
  .def("learn_string", &my_learn_string, "given an example specified as a string (as in a VW data file), learn on that example")
  .def("predict", &my_predict_string, "given a pyvw example, predict on that example")
  .def("predict_string", &my_predict_string, "given an example specified as a string (as in a VW data file), predict on that example")
  .def("hash_space", &VW::hash_space, "given a namespace (as a string), compute the hash of that namespace")
  .def("hash_feature", &VW::hash_feature, "given a feature string (arg2) and a hashed namespace (arg3), hash that feature")
  .def("finish_example", &my_finish_example, "tell VW that you're done with a given example")
  .def("setup_example", &my_setup_example, "given an example that you've created by hand, prepare it for learning (eg, compute quadratic feature)")
  .def("unsetup_example", &unsetup_example, "reverse the process of setup, so that you can go back and modify this example")

  .def("num_weights", &VW::num_weights, "how many weights are we learning?")
  .def("get_weight", &VW::get_weight, "get the weight for a particular index")
  .def("set_weight", &VW::set_weight, "set the weight for a particular index")
  .def("get_stride", &VW::get_stride, "return the internal stride")

  .def("get_sum_loss", &get_sum_loss, "return the total cumulative loss suffered so far")
  .def("get_weighted_examples", &get_weighted_examples, "return the total weight of examples so far")

  .def("get_search_ptr", &get_search_ptr, "return a pointer to the search data structure")
  .def("audit_example", &my_audit_example, "print example audit information")

  .def_readonly("lDefault", lDEFAULT, "Default label type (whatever vw was initialized with) -- used as input to the example() initializer")
  .def_readonly("lBinary", lBINARY, "Binary label type -- used as input to the example() initializer")
  .def_readonly("lMulticlass", lMULTICLASS, "Multiclass label type -- used as input to the example() initializer")
  .def_readonly("lCostSensitive", lCOST_SENSITIVE, "Cost sensitive label type (for LDF!) -- used as input to the example() initializer")
  .def_readonly("lContextualBandit", lCONTEXTUAL_BANDIT, "Contextual bandit label type -- used as input to the example() initializer")
  ;

  // define the example class
  py::class_<example, example_ptr>("example", py::no_init)
  .def("__init__", py::make_constructor(my_read_example), "Given a string as an argument parse that into a VW example (and run setup on it) -- default to multiclass label type")
  .def("__init__", py::make_constructor(my_empty_example), "Construct an empty (non setup) example; you must provide a label type (vw.lBinary, vw.lMulticlass, etc.)")

  .def("set_test_only", &my_set_test_only, "Change the test-only bit on an example")

  .def("get_tag", &my_get_tag, "Returns the tag associated with this example")
  .def("get_topic_prediction", &VW::get_topic_prediction, "For LDA models, returns the topic prediction for the topic id given")
  .def("get_feature_number", &VW::get_feature_number, "Returns the total number of features for this example")

  .def("get_example_counter", &get_example_counter, "Returns the counter of total number of examples seen up to and including this one")
  .def("get_ft_offset", &get_ft_offset, "Returns the feature offset for this example (used, eg, by multiclass classification to bulk offset all features)")
  .def("get_partial_prediction", &get_partial_prediction, "Returns the partial prediction associated with this example")
  .def("get_updated_prediction", &get_updated_prediction, "Returns the partial prediction as if we had updated it after learning")
  .def("get_loss", &get_loss, "Returns the loss associated with this example")
  .def("get_example_t", &get_example_t, "The total sum of importance weights up to and including this example")
  .def("get_total_sum_feat_sq", &get_total_sum_feat_sq, "The total sum of feature-value squared for this example")

  .def("num_namespaces", &ex_num_namespaces, "The total number of namespaces associated with this example")
  .def("namespace", &ex_namespace, "Get the namespace id for namespace i (for i = 0.. num_namespaces); specifically returns the ord() of the corresponding character id")
  .def("sum_feat_sq", &ex_sum_feat_sq, "Get the sum of feature-values squared for a given namespace id (id=character-ord)")
  .def("num_features_in", &ex_num_features, "Get the number of features in a given namespace id (id=character-ord)")
  .def("feature", &ex_feature, "Get the feature id for the ith feature in a given namespace id (id=character-ord)")
  .def("feature_weight", &ex_feature_weight, "The the feature value (weight) per .feature(...)")

  .def("push_hashed_feature", &ex_push_feature, "Add a hashed feature to a given namespace (id=character-ord)")
  .def("push_feature_list", &ex_push_feature_list, "Add a (Python) list of features to a given namespace")
  .def("push_feature_dict", &ex_push_dictionary, "Add a (Python) dictionary of namespace/feature-list pairs")
  .def("pop_feature", &ex_pop_feature, "Remove the top feature from a given namespace; returns True iff the list was non-empty")
  .def("push_namespace", &ex_push_namespace, "Add a new namespace")
  .def("ensure_namespace_exists", &ex_ensure_namespace_exists, "Add a new namespace if it doesn't already exist")
  .def("pop_namespace", &ex_pop_namespace, "Remove the top namespace off; returns True iff the list was non-empty")
  .def("erase_namespace", &ex_erase_namespace, "Remove all the features from a given namespace")

  .def("set_label_string", &ex_set_label_string, "(Re)assign the label of this example to this string")
  .def("get_simplelabel_label", &ex_get_simplelabel_label, "Assuming a simple_label label type, return the corresponding label (class/regression target/etc.)")
  .def("get_simplelabel_weight", &ex_get_simplelabel_weight, "Assuming a simple_label label type, return the importance weight")
  .def("get_simplelabel_initial", &ex_get_simplelabel_initial, "Assuming a simple_label label type, return the initial (baseline) prediction")
  .def("get_simplelabel_prediction", &ex_get_simplelabel_prediction, "Assuming a simple_label label type, return the final prediction")
  .def("get_multiclass_label", &ex_get_multiclass_label, "Assuming a multiclass label type, get the true label")
  .def("get_multiclass_weight", &ex_get_multiclass_weight, "Assuming a multiclass label type, get the importance weight")
  .def("get_multiclass_prediction", &ex_get_multiclass_prediction, "Assuming a multiclass label type, get the prediction")
  .def("get_costsensitive_prediction", &ex_get_costsensitive_prediction, "Assuming a cost_sensitive label type, get the prediction")
  .def("get_costsensitive_num_costs", &ex_get_costsensitive_num_costs, "Assuming a cost_sensitive label type, get the total number of label/cost pairs")
  .def("get_costsensitive_cost", &ex_get_costsensitive_cost, "Assuming a cost_sensitive label type, get the cost for a given pair (i=0.. get_costsensitive_num_costs)")
  .def("get_costsensitive_class", &ex_get_costsensitive_class, "Assuming a cost_sensitive label type, get the label for a given pair (i=0.. get_costsensitive_num_costs)")
  .def("get_costsensitive_partial_prediction", &ex_get_costsensitive_partial_prediction, "Assuming a cost_sensitive label type, get the partial prediction for a given pair (i=0.. get_costsensitive_num_costs)")
  .def("get_costsensitive_wap_value", &ex_get_costsensitive_wap_value, "Assuming a cost_sensitive label type, get the weighted-all-pairs recomputed cost for a given pair (i=0.. get_costsensitive_num_costs)")
  .def("get_cbandits_prediction", &ex_get_cbandits_prediction, "Assuming a contextual_bandits label type, get the prediction")
  .def("get_cbandits_num_costs", &ex_get_cbandits_num_costs, "Assuming a contextual_bandits label type, get the total number of label/cost pairs")
  .def("get_cbandits_cost", &ex_get_cbandits_cost, "Assuming a contextual_bandits label type, get the cost for a given pair (i=0.. get_cbandits_num_costs)")
  .def("get_cbandits_class", &ex_get_cbandits_class, "Assuming a contextual_bandits label type, get the label for a given pair (i=0.. get_cbandits_num_costs)")
  .def("get_cbandits_probability", &ex_get_cbandits_probability, "Assuming a contextual_bandits label type, get the bandits probability for a given pair (i=0.. get_cbandits_num_costs)")
  .def("get_cbandits_partial_prediction", &ex_get_cbandits_partial_prediction, "Assuming a contextual_bandits label type, get the partial prediction for a given pair (i=0.. get_cbandits_num_costs)")
  ;

  py::class_<Search::predictor, predictor_ptr>("predictor", py::no_init)
  .def("set_input", &my_set_input, "set the input (an example) for this predictor (non-LDF mode only)")
  //.def("set_input_ldf", &my_set_input_ldf, "set the inputs (a list of examples) for this predictor (LDF mode only)")
  .def("set_input_length", &Search::predictor::set_input_length, "declare the length of an LDF-sequence of examples")
  .def("set_input_at", &my_set_input_at, "put a given example at position in the LDF sequence (call after set_input_length)")
  .def("add_oracle", &my_add_oracle, "add an action to the current list of oracle actions")
  .def("add_oracles", &my_add_oracles, "add a list of actions to the current list of oracle actions")
  .def("add_allowed", &my_add_allowed, "add an action to the current list of allowed actions")
  .def("add_alloweds", &my_add_alloweds, "add a list of actions to the current list of allowed actions")
  .def("add_condition", &my_add_condition, "add a (tag,char) pair to the list of variables on which to condition")
  .def("add_condition_range", &my_add_condition_range, "given (tag,len,char), add (tag,char), (tag-1,char+1), ..., (tag-len,char+len) to the list of conditionings")
  .def("set_oracle", &my_set_oracle, "set an action as the current list of oracle actions")
  .def("set_oracles", &my_set_oracles, "set a list of actions as the current list of oracle actions")
  .def("set_allowed", &my_set_allowed, "set an action as the current list of allowed actions")
  .def("set_alloweds", &my_set_alloweds, "set a list of actions as the current list of allowed actions")
  .def("set_condition", &my_set_condition, "set a (tag,char) pair as the list of variables on which to condition")
  .def("set_condition_range", &my_set_condition_range, "given (tag,len,char), set (tag,char), (tag-1,char+1), ..., (tag-len,char+len) as the list of conditionings")
  .def("set_learner_id", &my_set_learner_id, "select the learner with which to make this prediction")
  .def("set_tag", &my_set_tag, "change the tag of this prediction")
  .def("predict", &Search::predictor::predict, "make a prediction")
  ;

  py::class_<Search::search, search_ptr>("search")
  .def("set_options", &Search::search::set_options, "Set global search options (auto conditioning, etc.)")
  .def("set_num_learners", &Search::search::set_num_learners, "Set the total number of learners you want to train")
  .def("get_history_length", &Search::search::get_history_length, "Get the value specified by --search_history_length")
  .def("loss", &Search::search::loss, "Declare a (possibly incremental) loss")
  .def("should_output", &search_should_output, "Check whether search wants us to output (only happens if you have -p running)")
  .def("predict_needs_example", &Search::search::predictNeedsExample, "Check whether a subsequent call to predict is actually going to use the example you pass---i.e., can you skip feature computation?")
  .def("output", &search_output, "Add a string to the coutput (should only do if should_output returns True)")
  .def("get_num_actions", &search_get_num_actions, "Return the total number of actions search was initialized with")
  .def("set_structured_predict_hook", &set_structured_predict_hook, "Set the hook (function pointer) that search should use for structured prediction (you don't want to call this yourself!")
  .def("set_force_oracle", &set_force_oracle, "For oracle decoding when .predict is run")
  .def("is_ldf", &Search::search::is_ldf, "check whether this search task is running in LDF mode")

  .def("po_exists", &po_exists, "For program (cmd line) options, check to see if a given option was specified; eg sch.po_exists(\"search\") should be True")
  .def("po_get", &po_get, "For program (cmd line) options, if an option was specified, get its value; eg sch.po_get(\"search\") should return the # of actions (returns either int or string)")
  .def("po_get_str", &po_get_string, "Same as po_get, but specialized for string return values.")
  .def("po_get_int", &po_get_int, "Same as po_get, but specialized for integer return values.")

  .def("get_predictor", &get_predictor, "Get a predictor object that can be used for making predictions; requires a tag argument to tag the prediction.")

  .def_readonly("AUTO_CONDITION_FEATURES", Search::AUTO_CONDITION_FEATURES, "Tell search to automatically add features based on conditioned-on variables")
  .def_readonly("AUTO_HAMMING_LOSS", Search::AUTO_HAMMING_LOSS, "Tell search to automatically compute hamming loss over predictions")
  .def_readonly("EXAMPLES_DONT_CHANGE", Search::EXAMPLES_DONT_CHANGE, "Tell search that on a single structured 'run', you don't change the examples you pass to predict")
  .def_readonly("IS_LDF", Search::IS_LDF, "Tell search that this is an LDF task")
  ;
}