dataset.h

Go to the documentation of this file.

 
#ifndef THEORETICA_DATASET_H
#define THEORETICA_DATASET_H
 
#ifndef THEORETICA_NO_PRINT
#include <sstream>
#include <ostream>
#endif
 
#include "./core_traits.h"
#include "./constants.h"
#include "./error.h"
 
 
namespace theoretica {
 

    template<typename Vector, enable_vector<Vector> = true>
    inline auto product(const Vector& X) {
 
        if(!X.size()) {
            TH_MATH_ERROR("product", X.size(), MathError::InvalidArgument);
            return vector_element_t<Vector>(nan());
        }
 
        auto res = X[0];
        for(unsigned int i = 1; i < X.size(); i++)
            res *= X[i];
 
        return res;
    }
 

    template<typename Vector>
    inline auto product_sum(const Vector& X, const Vector& Y) {
 
        if(X.size() != Y.size() || !X.size()) {
            TH_MATH_ERROR("product_sum", X.size(), MathError::InvalidArgument);
            return nan();
        }
 
        auto res = X[0] * Y[0];
        for(unsigned int i = 1; i < X.size(); i++)
            res += X[i] * Y[i];
 
        return res;
    }
 

    template<typename Vector>
    inline auto product_sum_squares(const Vector& X, const Vector& Y) {
 
        if(X.size() != Y.size() || !X.size()) {
            TH_MATH_ERROR("product_sum_squares", X.size(), MathError::InvalidArgument);
            return nan();
        }
 
        auto res = (X[0] * X[0]) * (Y[0] * Y[0]);
 
        for(unsigned int i = 1; i < X.size(); i++)
            res += (X[i] * X[i]) * (Y[i] * Y[i]);
 
        return res;
    }
 

    template<typename Vector>
    inline auto product_sum(const Vector& X, const Vector& Y, const Vector& Z) {
 
        if(X.size() != Y.size() || X.size() != Z.size() || !X.size()) {
            TH_MATH_ERROR("product_sum", X.size(), MathError::InvalidArgument);
            return nan();
        }
 
        auto res = X[0] * Y[0] * Z[0];
        for(unsigned int i = 1; i < X.size(); i++)
            res += X[i] * Y[i] * Z[i];
 
        return res;
    }
 

    template<typename Vector>
    inline auto quotient_sum(const Vector& X, const Vector& Y) {
 
        if(X.size() != Y.size()) {
            TH_MATH_ERROR("quotient_sum", X.size(), MathError::InvalidArgument);
            return nan();
        }
 
        if(abs(Y[0]) < MACH_EPSILON) {
            TH_MATH_ERROR("quotient_sum", Y[0], MathError::DivByZero);
            return nan();
        }
 
        auto res = X[0] / Y[0];
        for(unsigned int i = 1; i < X.size(); i++) {
 
            if(abs(Y[i]) < MACH_EPSILON) {
                TH_MATH_ERROR("quotient_sum", Y[i], MathError::DivByZero);
                return nan();
            }
 
            res += X[i] / Y[i];
        }
 
        return res;
    }
 

    template<typename Vector>
    inline auto sum_squares(const Vector& X) {
 
        if(!X.size()) {
            TH_MATH_ERROR("sum_squares", X.size(), MathError::InvalidArgument);
            return nan();
        }
 
        auto res = X[0] * X[0];
        for(unsigned int i = 1; i < X.size(); i++)
            res += X[i] * X[i];
 
        return res;
    }
 

    template<typename Vector>
    inline real sum_compensated(const Vector& X) {
 
        // Total sum
        real sum = 0;
 
        // Correction term
        real corr = 0;
 
        for (unsigned int i = 0; i < X.size(); i++) {
 
            const real temp = sum + X[i];
 
            // Sort the two addends to preserve bits
            corr += (abs(sum) >= abs(X[i]))
              ? ((sum - temp) + X[i])
              : ((X[i] - temp) + sum);
 
            sum = temp;
        }
 
        // Apply correction
        return sum + corr;
    }
 

    template<typename Vector>
    inline real sum_pairwise(
        const Vector& X, size_t begin = 0,
        size_t end = 0, size_t base_size = 128) {
 
        if(end == 0)
            end = X.size();
 
        real sum = 0;
 
        // Base case with given size (defaults to 128)
        if((end - begin) <= base_size) {
 
            for (size_t i = begin; i < end; ++i)
                sum += X[i];
 
        } else {
 
            // Recursive sum of two halves
            const size_t m = (end - begin) / 2;
            const size_t cutoff = begin + m;
 
            sum = sum_pairwise(X, begin, cutoff, base_size)
                + sum_pairwise(X, cutoff, end, base_size);
        }
 
        return sum;
    }
 

    template <
        typename Vector,
        std::enable_if_t<has_real_elements<Vector>::value> = true
    >
    inline auto sum(const Vector& X) {
        return sum_pairwise(X);
    }
 

    template<typename Vector>
    inline auto sum(const Vector& X) {
 
        // Use pairwise sum to reduce floating point errors.
        if TH_CONSTIF (has_real_elements<Vector>())
            return sum_pairwise(X);
 
        auto res = X[0];
        for(unsigned int i = 1; i < X.size(); i++)
            res += X[i];
 
        return res;
    }
 

    template<typename Vector, typename Function>
    inline Vector& transform(Function f, Vector& X) {
 
        for (unsigned int i = 0; i < X.size(); i++)
            X[i] = f(X[i]);
 
        return X;
    }
 

    template<typename Vector1, typename Vector2 = Vector1, typename Function>
    inline Vector2& map(Function f, const Vector1& src, Vector2& dest) {
 
        if(src.size() != dest.size()) {
            TH_MATH_ERROR("th::map", dest.size(), MathError::InvalidArgument);
            return algebra::vec_error(dest);
        }
 
        for (unsigned int i = 0; i < src.size(); i++)
            dest[i] = f(src[i]);
 
        return dest;
    }
 

    template<typename Vector2, typename Vector1, typename Function>
    inline Vector2 map(Function f, const Vector1& X) {
 
        Vector2 res;
        res.resize(X.size());
 
        for (unsigned int i = 0; i < X.size(); i++)
            res[i] = f(X[i]);
 
        return res;
    }
 

    template<typename Vector, typename Function>
    inline Vector map(Function f, const Vector& X) {
        return map<Vector, Vector, Function>(f, X);
    }
 

    template<typename Vector1, typename Vector2, typename Vector3 = Vector1>
    inline Vector3 concatenate(const Vector1& v1, const Vector2& v2) {
 
        Vector3 res;
        res.resize(v1.size() + v2.size());
        const unsigned int offset = v1.size();
 
        for (unsigned int i = 0; i < offset; ++i)
            res[i] = v1[i];
 
        for (unsigned int i = 0; i < v2.size(); ++i)
            res[i + offset] = v2[i];
 
        return res;
    }
 

    template<typename Vector>
    inline auto max(const Vector& X) {
 
        using Type = vector_element_t<Vector>;
 
        if(!X.size()) {
            TH_MATH_ERROR("max", X.size(), MathError::InvalidArgument);
            return make_error<Type>();
        }
 
        auto curr = X[0];
 
        for (unsigned int i = 1; i < X.size(); ++i)
            if(X[i] > curr)
                curr = X[i];
 
        return curr;
    }
 

    template<typename Vector>
    inline auto min(const Vector& X) {
 
        using Type = vector_element_t<Vector>;
 
        if(!X.size()) {
            TH_MATH_ERROR("min", X.size(), MathError::InvalidArgument);
            return make_error<Type>();
        }
 
        auto curr = X[0];
 
        for (unsigned int i = 1; i < X.size(); ++i)
            if(X[i] < curr)
                curr = X[i];
 
        return curr;
    }
 
 
    // Different types of means
 

    template<typename Dataset>
    inline real arithmetic_mean(const Dataset& data) {
 
        if(!data.size()) {
            TH_MATH_ERROR("arithmetic_mean", data.size(), MathError::DivByZero);
            return nan();
        }
 
        // Sum of x_i / N
        return sum(data) / (real) data.size();
    }
 

    template<typename Dataset>
    inline real harmonic_mean(const Dataset& data) {
 
        if(!data.size()) {
            TH_MATH_ERROR("harmonic_mean", data.size(), MathError::DivByZero);
            return nan();
        }
 
        real res = 0;
 
        for (unsigned int i = 0; i < data.size(); ++i) {
 
            if(data[i] == 0) {
                TH_MATH_ERROR("harmonic_mean", data[i], MathError::DivByZero);
                return nan();
            }
 
            res += 1.0 / data[i];
        }
 
        return static_cast<real>(data.size()) / res;
    }
 

    template<typename Dataset>
    inline real geometric_mean(const Dataset& data) {
        return root(product(data), data.size());
    }
 

    template<typename Dataset1, typename Dataset2>
    inline real weighted_mean(const Dataset1& data, const Dataset2& weights) {
 
        // Sum of x_i * w_i / Sum of w_i
        return product_sum(data, weights) / sum(weights);
    }
 

    template<typename Dataset>
    inline real quadratic_mean(const Dataset& data) {
 
        if(!data.size()) {
            TH_MATH_ERROR("quadratic_mean", data.size(), MathError::InvalidArgument);
            return nan();
        }
 
        return sqrt(sum_squares(data) / data.size());
    }
}
 
#endif