/******************************************
Copyright (c) 2016, Yuri Malitsky and Horst Samulowitz
Copyright (c) 2018, Mate Soos

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
***********************************************/

#include <vector>
#include <cmath>

#include "solver.h"
#include "features_calc.h"

using std::vector;
using namespace CMSat;

template<class Function, class Function2>
void SolveFeaturesCalc::for_one_clause(
    const Watched& cl
    , const Lit lit
    ,  Function func_each_cl
    ,  Function2 func_each_lit
) const {
    unsigned neg_vars = 0;
    unsigned pos_vars = 0;
    unsigned size = 0;

    switch (cl.getType()) {
        case CMSat::watch_binary_t: {
            if (cl.red()) {
                //only irred cls
                break;
            }
            if (lit > cl.lit2()) {
                //only count once
                break;
            }

            pos_vars += !lit.sign();
            pos_vars += !cl.lit2().sign();
            size = 2;
            neg_vars = size - pos_vars;
            func_each_cl(size, pos_vars, neg_vars);
            func_each_lit(lit, size, pos_vars, neg_vars);
            func_each_lit(cl.lit2(), size, pos_vars, neg_vars);
            break;
        }

        case CMSat::watch_clause_t: {
            const Clause& clause = *solver->cl_alloc.ptr(cl.get_offset());
            if (clause.red()) {
                //only irred cls
                break;
            }
            if (clause[0] < clause[1]) {
                //only count once
                break;
            }

            for (const Lit cl_lit : clause) {
                pos_vars += !cl_lit.sign();
            }
            size = clause.size();
            neg_vars = size - pos_vars;
            func_each_cl(size, pos_vars, neg_vars);
            for (const Lit cl_lit : clause) {
                func_each_lit(cl_lit, size, pos_vars, neg_vars);
            }
            break;
        }

        case CMSat::watch_idx_t: {
             // This should never be here
            assert(false);
            exit(-1);
            break;
        }
    }
}

template<class Function, class Function2>
void SolveFeaturesCalc::for_all_clauses(Function func_each_cl, Function2 func_each_lit) const
{
    for (size_t i = 0; i < solver->nVars() * 2; i++) {
        Lit lit = Lit::toLit(i);
        for (const Watched & w : solver->watches[lit]) {
            for_one_clause(w, lit, func_each_cl, func_each_lit);
        }
    }
}

void SolveFeaturesCalc::fill_vars_cls()
{
    feat.numVars = solver->nVars();
    feat.numClauses = solver->longIrredCls.size() + solver->binTri.irredBins;
    myVars.resize(solver->nVars());

    auto func_each_cl = [&](unsigned /*size*/, unsigned pos_vars, unsigned /*neg_vars*/) -> bool {
        if (pos_vars <= 1 ) {
            feat.horn += 1;
            return true;
        }
        return false;
    };
    auto func_each_lit = [&](Lit lit, unsigned /*size*/, unsigned pos_vars, unsigned /*neg_vars*/) -> void {
        if (pos_vars <= 1 ) {
            myVars[lit.var()].horn++;
        }

        if (!lit.sign()) {
            myVars[lit.var()].numPos++;
        }
        myVars[lit.var()].size++;
    };
    for_all_clauses(func_each_cl, func_each_lit);
}

void SolveFeaturesCalc::calculate_clause_stats()
{
    auto empty_func = [](const Lit, unsigned /*size*/, unsigned /*pos_vars*/, unsigned /*neg_vars*/) -> void {};
    auto func_each_cl = [&](unsigned size, unsigned pos_vars, unsigned /*neg_vars*/) -> void {
        if (size == 0 ) {
            return;
        }

        double _size = (double)size / (double)feat.numVars;
        feat.vcg_cls_min = std::min(feat.vcg_cls_min, _size);
        feat.vcg_cls_max = std::max(feat.vcg_cls_max, _size);
        feat.vcg_cls_mean += _size;

        double _pnr = 0.5 + ((2.0 * (double)pos_vars - (double)size) / (2.0 * (double)size));
        feat.pnr_cls_min = std::min(feat.pnr_cls_min, _pnr);
        feat.pnr_cls_max = std::max(feat.pnr_cls_max, _pnr);
        feat.pnr_cls_mean += _pnr;
    };
    for_all_clauses(func_each_cl, empty_func);

    feat.vcg_cls_mean /= (double)feat.numClauses;
    feat.pnr_cls_mean /= (double)feat.numClauses;
    feat.horn /= (double)feat.numClauses;
    feat.binary = float_div(solver->binTri.irredBins, feat.numClauses);

    feat.vcg_cls_spread = feat.vcg_cls_max - feat.vcg_cls_min;
    feat.pnr_cls_spread = feat.pnr_cls_max - feat.pnr_cls_min;
}

void SolveFeaturesCalc::calculate_variable_stats()
{
    if (feat.numVars == 0)
        return;

    for ( int vv = 0; vv < (int)myVars.size(); vv++ ) {
        if ( myVars[vv].size == 0 ) {
            continue;
        }

        double _size = myVars[vv].size / (double)feat.numClauses;
        feat.vcg_var_min = std::min(feat.vcg_var_min, _size);
        feat.vcg_var_max = std::max(feat.vcg_var_max, _size);
        feat.vcg_var_mean += _size;

        double _pnr = 0.5 + ((2.0 * myVars[vv].numPos - myVars[vv].size)
                             / (2.0 * myVars[vv].size));
        feat.pnr_var_min = std::min(feat.pnr_var_min, _pnr);
        feat.pnr_var_max = std::max(feat.pnr_var_max, _pnr);
        feat.pnr_var_mean += _pnr;

        double _horn = myVars[vv].horn / (double)feat.numClauses;
        feat.horn_min = std::min(feat.horn_min, _horn);
        feat.horn_max = std::max(feat.horn_max, _horn);
        feat.horn_mean += _horn;
    }

    if (feat.vcg_var_mean > 0) {
        feat.vcg_var_mean /= (double)feat.numVars;
    }
    if (feat.pnr_var_mean > 0) {
        feat.pnr_var_mean /= (double)feat.numVars;
    }
    if (feat.horn_mean > 0) {
        feat.horn_mean /= (double)feat.numVars;
    }

    feat.vcg_var_spread = feat.vcg_var_max - feat.vcg_var_min;
    feat.pnr_var_spread = feat.pnr_var_max - feat.pnr_var_min;
    feat.horn_spread = feat.horn_max - feat.horn_min;
}

void SolveFeaturesCalc::calculate_extra_clause_stats()
{
    auto empty_func = [](const Lit, unsigned /*size*/, unsigned /*pos_vars*/, unsigned /*neg_vars*/) -> void {};
    auto each_clause = [&](unsigned size, unsigned pos_vars, unsigned /*neg_vars*/) -> void {
        if ( size == 0 ) {
            return;
        }

        double _size = (double)size / (double)feat.numVars;
        feat.vcg_cls_std += (feat.vcg_cls_mean - _size) * (feat.vcg_cls_mean - _size);

        double _pnr = 0.5 + ((2.0 * (double)pos_vars - (double)size) / (2.0 * (double)size));
        feat.pnr_cls_std += (feat.pnr_cls_mean - _pnr) * (feat.pnr_cls_mean - _pnr);
    };
    for_all_clauses(each_clause, empty_func);

    if ( feat.vcg_cls_std > feat.eps && feat.vcg_cls_mean > feat.eps ) {
        feat.vcg_cls_std = std::sqrt(feat.vcg_cls_std / (double)feat.numClauses) / feat.vcg_cls_mean;
    } else {
        feat.vcg_cls_std = 0;
    }
    if ( feat.pnr_cls_std > feat.eps && feat.pnr_cls_mean > feat.eps ) {
        feat.pnr_cls_std = std::sqrt(feat.pnr_cls_std / (double)feat.numClauses) / feat.pnr_cls_mean;
    } else {
        feat.pnr_cls_std = 0;
    }
}

void SolveFeaturesCalc::calculate_extra_var_stats()
{
    if (feat.numVars == 0)
        return;

    for ( int vv = 0; vv < (int)myVars.size(); vv++ ) {
        if ( myVars[vv].size == 0 ) {
            continue;
        }

        double _size = myVars[vv].size / (double)feat.numClauses;
        feat.vcg_var_std += (feat.vcg_var_mean - _size) * (feat.vcg_var_mean - _size);

        double _pnr = 0.5 + ((2.0 * myVars[vv].numPos - myVars[vv].size) / (2.0 * myVars[vv].size));
        feat.pnr_var_std += (feat.pnr_var_mean - _pnr) * (feat.pnr_var_mean - _pnr);

        double _horn = myVars[vv].horn / (double)feat.numClauses;
        feat.horn_std += (feat.horn_mean - _horn) * (feat.horn_mean - _horn);
    }
    if ( feat.vcg_var_std > feat.eps && feat.vcg_var_mean > feat.eps ) {
        feat.vcg_var_std = std::sqrt(feat.vcg_var_std / (double)feat.numVars) / feat.vcg_var_mean;
    } else {
        feat.vcg_var_std = 0;
    }

    if ( feat.pnr_var_std > feat.eps && feat.pnr_var_mean > feat.eps
        && feat.pnr_var_mean != 0
    ) {
        feat.pnr_var_std = std::sqrt(feat.pnr_var_std / (double)feat.numVars) / feat.pnr_var_mean;
    } else {
        feat.pnr_var_std = 0;
    }

    if ( feat.horn_std / (double)feat.numVars > feat.eps && feat.horn_mean > feat.eps
        && feat.horn_mean != 0
    ) {
        feat.horn_std = std::sqrt(feat.horn_std / (double)feat.numVars) / feat.horn_mean;
    } else {
        feat.horn_std = 0;
    }
}

void SolveFeaturesCalc::calculate_cl_distributions(
    const vector<ClOffset>& clauses
    , struct SolveFeatures::Distrib& distrib_data
) {
    if (clauses.empty()) {
        return;
    }

    double glue_mean = 0;
    double glue_var = 0;

    double size_mean = 0;
    double size_var = 0;

    double activity_mean = 0;
    double activity_var = 0;

    //Calculate means
    double cla_inc = solver->get_cla_inc();
    for(ClOffset off: clauses)
    {
        const Clause& cl = *solver->cl_alloc.ptr(off);
        size_mean += cl.size();
        glue_mean += cl.stats.glue;
        if (cl.red()) {
            activity_mean += (double)cl.stats.activity/cla_inc;
        }
    }
    size_mean /= clauses.size();
    glue_mean /= clauses.size();
    activity_mean /= clauses.size();

    //Calculate variances
    for(ClOffset off: clauses)
    {
        const Clause& cl = *solver->cl_alloc.ptr(off);
        size_var += std::pow(size_mean-cl.size(), 2);
        glue_var += std::pow(glue_mean-cl.stats.glue, 2);
        activity_var += std::pow(activity_mean-(double)cl.stats.activity/cla_inc, 2);
    }
    size_var /= clauses.size();
    glue_var /= clauses.size();
    activity_var /= clauses.size();

    //Assign calculated values
    distrib_data.glue_distr_mean = glue_mean;
    distrib_data.glue_distr_var = glue_var;
    distrib_data.size_distr_mean = size_mean;
    distrib_data.size_distr_var = size_var;
    distrib_data.activity_distr_mean = activity_mean;
    distrib_data.activity_distr_var = activity_var;
}

void SolveFeaturesCalc::normalise_values()
{
    if (feat.vcg_var_min == std::numeric_limits<double>::max())
        feat.vcg_var_min = -1;
    if (feat.vcg_var_max == std::numeric_limits<double>::min())
        feat.vcg_var_max = -1;

    if (feat.vcg_cls_min  == std::numeric_limits<double>::max())
        feat.vcg_cls_min = -1;
    if (feat.vcg_cls_max == std::numeric_limits<double>::min())
        feat.vcg_cls_max = -1;

    if (feat.pnr_var_min == std::numeric_limits<double>::max())
        feat.pnr_var_min = -1;
    if (feat.pnr_var_max == std::numeric_limits<double>::min())
        feat.pnr_var_max = -1;

    if (feat.horn_min == std::numeric_limits<double>::max())
        feat.horn_min = -1;
    if (feat.horn_max == std::numeric_limits<double>::min())
        feat.horn_max = -1;

    if (feat.pnr_cls_min == std::numeric_limits<double>::max())
        feat.pnr_cls_min = -1;
    if (feat.pnr_cls_max == std::numeric_limits<double>::min())
        feat.pnr_cls_max = -1;
}

SolveFeatures SolveFeaturesCalc::extract()
{
    double start_time = cpuTime();
    fill_vars_cls();

    feat.numVars = 0;
    for ( int vv = 0; vv < (int)myVars.size(); vv++ ) {
        if ( myVars[vv].size > 0 ) {
            feat.numVars++;
        }
    }
    if (feat.numVars > 0) {
        feat.var_cl_ratio = (double)feat.numVars/ (double)feat.numClauses;
    }

    if (feat.numClauses > 0 && feat.numVars > 0) {
        calculate_clause_stats();
        calculate_variable_stats();

        calculate_extra_clause_stats();
        calculate_extra_var_stats();

        calculate_cl_distributions(solver->longRedCls[0], feat.red_cl_distrib);
        calculate_cl_distributions(solver->longIrredCls, feat.irred_cl_distrib);
    }
    normalise_values();

    if (solver->conf.verbosity > 5) {
        cout << "c [features] extracted"
        << solver->conf.print_times(cpuTime() - start_time)
        << endl;
    }

    return feat;
}