Tensor template class created

Author: Anthony

include/cslib/data_structure/tensor.hpp

@@ -0,0 +1,304 @@
+#ifndef CSLIB_DATA_STRUCTURE_TENSOR_HPP
+#define CSLIB_DATA_STRUCTURE_TENSOR_HPP
+
+#include <cstdint>
+#include <string>
+#include <vector>
+#include <numeric>
+
+namespace cslib {
+namespace data_structure
+{
+    /* Class representing tensor object.
+     * TODO: 
+     *  -broadcasting
+     *  -range cutting
+     *  -linear algebra operations
+     *  -dimension operations (expand, reshape, ...)
+     */
+    
+
+    template <typename T>
+    class Tensor
+    {
+    public:
+        using dimensions = std::vector<size_t>;
+
+    private:
+        T* _data = nullptr;
+        int _number_elements;
+        dimensions _dims;
+
+        bool _is_shallow_copy;
+        bool _is_sub_tensor;
+
+    public:
+        // Constructors
+        Tensor(const dimensions& dims) :
+            _is_shallow_copy(false),
+            _is_sub_tensor(false),
+            _dims(dims)
+        {
+            _number_elements = std::accumulate(_dims.begin(), _dims.end(), 1, std::multiplies<size_t>());
+            _data = new T[_number_elements];
+        }
+
+        Tensor(const Tensor& rhs)
+        {
+            data_release();
+
+            _data = rhs._data;
+            _number_elements = rhs._number_elements;
+            _dims = rhs._dims;
+            _is_sub_tensor = rhs._is_sub_tensor;
+            _is_shallow_copy = true;
+        }
+
+        Tensor(Tensor&& rhs) noexcept
+        {
+            data_release();
+
+            _data = rhs._data;
+            _number_elements = rhs._number_elements;
+            _dims = rhs._dims;
+            _is_sub_tensor = rhs._is_sub_tensor;
+            _is_shallow_copy = rhs._is_shallow_copy;
+
+            rhs._data = nullptr;
+            rhs._number_elements = 0;
+            rhs._dims = {};
+            rhs._is_shallow_copy = true;
+            rhs._is_sub_tensor = false;
+        }
+
+        // Destructor
+        ~Tensor() noexcept
+        {
+            data_release();
+        }
+
+        // Operators overloading
+        Tensor& operator=(const Tensor& rhs)
+        {
+            if (_is_shallow_copy && _is_sub_tensor) // copying sub tensor
+            {
+                // assert (_dims == rhs._dims)?
+                if (_dims == rhs._dims)
+                    for (auto i = 0; i < _number_elements; i++)
+                        _data[i] = rhs._data[i];
+            }
+            else // tensor is not sub tensor
+            {
+                data_release();
+
+                _data = rhs._data;
+                _number_elements = rhs._number_elements;
+                _dims = rhs._dims;
+                _is_shallow_copy = true;
+            }
+
+            return *this;
+        }
+
+        Tensor& operator=(Tensor&& rhs) noexcept
+        {
+            if (_is_shallow_copy && _is_sub_tensor) // copying sub tensor
+            {
+                // assert (_dims == rhs._dims)?
+                if (_dims == rhs._dims)
+                    for (auto i = 0; i < _number_elements; i++)
+                        _data[i] = rhs._data[i];
+            }
+            else // tensor is not sub tensor
+            {
+                data_release();
+
+                _data = rhs._data;
+                _number_elements = rhs._number_elements;
+                _dims = rhs._dims;
+                _is_shallow_copy = true;
+            }
+
+            rhs._data = nullptr;
+            rhs._number_elements = 0;
+            rhs._dims = {};
+            rhs._is_shallow_copy = true;
+            rhs._is_sub_tensor = false;
+
+            return *this;
+        }
+
+        Tensor& operator=(T value)
+        {
+            for (auto i = 0; i < _number_elements; i++)
+                _data[i] = value;
+
+            return *this;
+        }
+
+        Tensor operator+(T value) const
+        {
+            Tensor result = make_copy();
+
+            for (auto i = 0; i < _number_elements; i++)
+                result._data[i] = _data[i] + value;
+
+            return result;
+        }
+
+        Tensor operator+(const Tensor& rhs) const
+        {
+            Tensor result = make_copy();
+
+            if (rhs._dims == _dims) // element-wise addition
+            {
+                for (auto i = 0; i < _number_elements; i++)
+                    result._data[i] = _data[i] + rhs._data[i];
+            }
+
+            return result;
+        }
+
+        Tensor operator*(T value) const
+        {
+            Tensor result = make_copy();
+
+            for (auto i = 0; i < _number_elements; i++)
+                result._data[i] = _data[i] * value;
+
+            return result;
+        }
+
+        Tensor operator*(const Tensor& rhs) const
+        {
+            Tensor result = make_copy();
+
+            if (rhs._dims == _dims) // element-wise multiply
+            {
+                for (auto i = 0; i < _number_elements; i++)
+                    result._data[i] = _data[i] * rhs._data[i];
+            }
+
+            return result;
+        }
+
+        Tensor operator[](int index)
+        {
+            auto new_dimensions = _dims;
+            new_dimensions.erase(new_dimensions.begin());
+
+            if (new_dimensions.size() == 0)
+                new_dimensions.push_back(1);
+
+            auto data_offset = _data + index * std::accumulate(new_dimensions.begin(), new_dimensions.end(), 1, std::multiplies<size_t>());
+            return Tensor(new_dimensions, data_offset);
+        }
+
+        const Tensor operator[](int index) const
+        {
+            auto new_dimensions = _dims;
+            new_dimensions.erase(new_dimensions.begin());
+
+            if (new_dimensions.size() == 0)
+                new_dimensions.push_back(1);
+
+            auto data_offset = _data + index * std::accumulate(new_dimensions.begin(), new_dimensions.end(), 1, std::multiplies<size_t>());
+            return Tensor(new_dimensions, data_offset);
+        }
+
+        // deep copy operations
+        Tensor make_copy() const
+        {
+            auto copied_tensor = Tensor(_dims);
+            for (auto i = 0; i < _number_elements; i++)
+                copied_tensor._data[i] = _data[i];
+
+            return copied_tensor;
+        }
+
+        // Linear algebra
+        void transpose() 
+        {
+            auto tmp = _dims[0];
+            _dims[0] = _dims[1];
+            _dims[1] = tmp;
+
+            // in-place sort or some iterators magic requiered
+        }
+
+        // String representation
+        std::string to_string() const
+        {
+            std::string result = "";
+            auto num_dims = _dims.size();
+
+            if (num_dims > 2)
+            {
+                for (auto i = 0; i < _dims[0]; i++)
+                    result += (*this)[i].to_string();
+            }
+            else if (num_dims == 2)
+            {
+                auto _h = _dims[0];
+                auto _w = _dims[1];
+
+                result += "[";
+                for (auto i = 0; i < _number_elements; i += _w)
+                {
+                    if (i != 0)
+                        result += " [";
+                    else
+                        result += "[";
+
+                    for (auto j = 0; j < _w; j++)
+                    {
+                        if (j != (_w - 1))
+                            result += std::to_string(_data[i+j]) + ", ";
+                        else
+                            result += std::to_string(_data[i+j]);
+                    }
+                    result += "]";
+
+                    if (i != (_h - 1))
+                        result += "\n";
+                }
+                result += "]\n\n";
+            }
+            else
+            {
+                result += "[";
+                for (auto i = 0; i < _dims[0]; i++)
+                {
+                    if (i != (_dims[0] - 1))
+                        result += std::to_string(_data[i]) + ", ";
+                    else
+                        result += std::to_string(_data[i]);
+                }
+                result += "]  ";
+            }
+
+            return result.substr(0, result.length() - 2);
+        }
+
+    private:
+        // sub-tensor ctor
+        Tensor(const dimensions& dims, T* data_start_address)
+            : _is_shallow_copy(true),
+            _is_sub_tensor(true),
+            _data(data_start_address),
+            _dims(dims)
+        {
+            _number_elements = std::accumulate(_dims.begin(), _dims.end(), 1, std::multiplies<size_t>());
+        }
+
+        // Memory management
+        void data_release()
+        {
+            if (!_is_shallow_copy)
+                delete[] _data;
+        }
+    };
+
+}}
+
+#endif // CSLIB_DATA_STRUCTURE_TENSOR_HPP