C++

Forum

 
building a multilayer perceptron

closed account (E093605o)
Hello,

I am trying to build a MLP to learn C++. So far I have implemented a small linear algebra library "Matrix.h" and "Matrix.cpp". Now, I want to implement a feedforward function in the MLP.cpp file. However, I am stuck because I dont know how to build a function that applies a function to a Matrix parametrized with a type parameter. In line 188 in my code is the function of interest!

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
 
//Matrix.h

//NOTE: the part of interest is in the apply_function method all the way down!!
#pragma once
#include <vector>
#include <cmath>
#include <cassert>
#include <iostream>
#include <tuple>
#include <random>
#include <functional>

template<typename Type>
class Matrix {

  size_t cols{};
  size_t rows{};

 public:
  std::vector<std::vector<Type>> data;
  std::tuple<size_t, size_t> shape;
  size_t elementCount{};

  /* constructors */
  Matrix(size_t rowsArg, size_t colsArg) : cols(colsArg), rows(rowsArg),
  elementCount(rows*cols), shape(std::tuple<size_t,size_t>(rows,cols))
  {
		data = std::vector<std::vector<Type>>(rows,std::vector<Type>(cols));
	}

  Matrix(){};

  //methods
  void print();

  Matrix<Type> matmul(Matrix<Type> &m);

  Matrix<Type> multiply_elementwise(Matrix<Type> &m);
  
  Matrix<Type> multiply_scalar(Type scalar);

  Matrix<Type> square();

  Matrix<Type> add(Matrix<Type> &m);

  Matrix<Type> sub(Matrix &target);

  Matrix<Type> T();

  Matrix<Type> apply_function(Type (*func)(Type));

  Type& operator()(size_t row, size_t col) {
    assert(row < data.size() && col < data[0].size());
    return data[row][col];
}

 Matrix operator+(Matrix &target) {
    return add(target);
  }

  Matrix operator-() {
    Matrix output(rows, cols);
    for (size_t r = 0; r < rows; ++r) {
      for (size_t c = 0; c < cols; ++c) {
        output(r, c) = -(*this)(r, c);
      }
    }
    return output;
  } 

   Matrix operator-(Matrix &target) {  // for cleaner usage
    return sub(target);
  }
  

};

// methods
template<typename Type>
void Matrix<Type>::print(){
  for (int i = 0; i < rows; i++){
    for (int j = 0; j < cols; j++){
      std::cout << data[i][j] << " ";
    }
    std::cout << std::endl;
  }
}

template <typename Type>
Matrix<Type> Matrix<Type>::matmul(Matrix<Type> &target) {
    assert(cols == target.rows);
    Matrix output(rows, target.cols);

    for (size_t r = 0; r < output.rows; ++r) {
      for (size_t c = 0; c < output.cols; ++c) {
        for (size_t k = 0; k < target.rows; ++k)
          output(r, c) += (*this)(r, k) * target(k, c);
      }
    }
    return output;
  };

template <typename T>
struct mtx {
  static Matrix<T> randn(size_t rows, size_t cols) {
    Matrix<T> M(rows, cols);

    std::random_device rd{};
    std::mt19937 gen{rd()};

    // init Gaussian distr. w/ N(mean=0, stdev=1/sqrt(numel))
    T n(M.elementCount);
    T stdev{1 / sqrt(n)};
    std::normal_distribution<T> d{0, stdev};

    // fill each element w/ draw from distribution
    for (size_t r = 0; r < rows; ++r) {
      for (int c = 0; c < cols; ++c) {
        M(r, c) = d(gen);
      }
    }
    return M;
  }
};

template <typename Type>
Matrix<Type> Matrix<Type>::multiply_elementwise(Matrix<Type> &target){
    assert(shape == target.shape);
    Matrix output((*this));
    for (size_t r = 0; r < output.rows; ++r) {
      for (size_t c = 0; c < output.cols; ++c) {
        output(r, c) = target(r,c) * (*this)(r, c);
      }
    }
    return output;
  }

  template<typename Type>
  Matrix<Type> Matrix<Type>::square() { 
    Matrix output((*this));
    output = multiply_elementwise(output);
    return output;
  }

  template<typename Type>
  Matrix<Type> Matrix<Type>::multiply_scalar(Type scalar) {
    Matrix output((*this));
    for (size_t r = 0; r < output.rows; ++r) {
      for (size_t c = 0; c < output.cols; ++c) {
        output(r, c) = scalar * (*this)(r, c);
      }
    }
    return output;
  }

  template<typename Type>
  Matrix<Type> Matrix<Type>::add(Matrix &target) {
    assert(shape == target.shape);
    Matrix output(rows, std::get<1>(target.shape));

    for (size_t r = 0; r < output.rows; ++r) {
      for (size_t c = 0; c < output.cols; ++c) {
        output(r, c) = (*this)(r, c) + target(r, c);
      }
    }
    return output;
  }

  template<typename Type>
  Matrix<Type> Matrix<Type>::sub(Matrix &target) {
    Matrix neg_target = -target;
    return add(neg_target);
  }

  template<typename Type>
   Matrix<Type> Matrix<Type>::T() {
    size_t new_rows{cols}, new_cols{rows};
    Matrix transposed(new_rows, new_cols);

    for (size_t r = 0; r < new_rows; ++r) {
      for (size_t c = 0; c < new_cols; ++c) {
        transposed(r, c) = (*this)(c, r);  // swap row and col
      }
    }
    return transposed;
  }

  template<typename Type>
  Matrix<Type> Matrix<Type>::apply_function(Type (*func)(Type)) {
    Matrix output((*this));
    for (size_t r = 0; r < rows; ++r) {
      for (size_t c = 0; c < cols; ++c) {
        output(r, c) = function((*this)(r, c));
      }
    }
    return output;
  }

//MLP.cpp

#include "Matrix.h"
#include "MLP.h"
#include <iostream>


    void MLP::printParameters(){
        for(int i = 0; i < weights.size(); i++){
            std::cout << "weights in layer " << i << ":" << std::endl;
            weights[i].print();
            std::cout << "biases in layer " << i << ":" << std::endl;
            biases[i].print();
        }
    }
    inline double MLP::sigmoid(double x) {
        return 1.0 / (1 + exp(-x));
    }

/* here, passing sigmoid gives an error: argument of type 
   "double (MLP::*)(double x)" is incompatible with parameter of type 
"double (*)(double)" */
    void MLP::feedforward(Matrix<double> x){
        for (size_t i = 0; i < layers.size(); i++){
            auto z = weights[i].matmul(x) + biases[i];
            activations[i] = z.apply_function(sigmoid);
        }
    }
Last edited on
seeplus (6457)
This compiles. The main issue is that sigmoid needs to be defined as static. As you didn't provide compilable code, I've guessed a MLP struct that enables a compilation:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
 
#include <vector>
#include <cmath>
#include <cassert>
#include <iostream>
#include <tuple>
#include <random>
#include <functional>

template<typename Type>
class Matrix {

	size_t cols {};
	size_t rows {};

public:
	std::vector<std::vector<Type>> data;
	std::tuple<size_t, size_t> shape;
	size_t elementCount {};

	/* constructors */
	Matrix(size_t rowsArg, size_t colsArg) : cols(colsArg), rows(rowsArg),
		elementCount(rows* cols), shape(std::tuple<size_t, size_t>(rows, cols)) {
		data = std::vector<std::vector<Type>>(rows, std::vector<Type>(cols));
	}

	Matrix() {};

	//methods
	void print();

	Matrix<Type> matmul(Matrix<Type>& m);

	Matrix<Type> multiply_elementwise(Matrix<Type>& m);

	Matrix<Type> multiply_scalar(Type scalar);

	Matrix<Type> square();

	Matrix<Type> add(Matrix<Type>& m);

	Matrix<Type> sub(Matrix& target);

	Matrix<Type> T();

	Matrix<Type> apply_function(Type(*func)(Type));

	Type& operator()(size_t row, size_t col) {
		assert(row < data.size() && col < data[0].size());
		return data[row][col];
	}

	Matrix operator+(Matrix& target) {
		return add(target);
	}

	Matrix operator-() {
		Matrix output(rows, cols);
		for (size_t r = 0; r < rows; ++r) {
			for (size_t c = 0; c < cols; ++c) {
				output(r, c) = -(*this)(r, c);
			}
		}
		return output;
	}

	Matrix operator-(Matrix& target) {  // for cleaner usage
		return sub(target);
	}


};

// methods
template<typename Type>
void Matrix<Type>::print() {
	for (int i = 0; i < rows; i++) {
		for (int j = 0; j < cols; j++) {
			std::cout << data[i][j] << " ";
		}
		std::cout << std::endl;
	}
}

template <typename Type>
Matrix<Type> Matrix<Type>::matmul(Matrix<Type>& target) {
	assert(cols == target.rows);
	Matrix output(rows, target.cols);

	for (size_t r = 0; r < output.rows; ++r) {
		for (size_t c = 0; c < output.cols; ++c) {
			for (size_t k = 0; k < target.rows; ++k)
				output(r, c) += (*this)(r, k) * target(k, c);
		}
	}
	return output;
};

template <typename T>
struct mtx {
	static Matrix<T> randn(size_t rows, size_t cols) {
		Matrix<T> M(rows, cols);

		std::random_device rd {};
		std::mt19937 gen { rd() };

		// init Gaussian distr. w/ N(mean=0, stdev=1/sqrt(numel))
		T n(M.elementCount);
		T stdev { 1 / sqrt(n) };
		std::normal_distribution<T> d { 0, stdev };

		// fill each element w/ draw from distribution
		for (size_t r = 0; r < rows; ++r) {
			for (int c = 0; c < cols; ++c) {
				M(r, c) = d(gen);
			}
		}
		return M;
	}
};

template <typename Type>
Matrix<Type> Matrix<Type>::multiply_elementwise(Matrix<Type>& target) {
	assert(shape == target.shape);
	Matrix output((*this));
	for (size_t r = 0; r < output.rows; ++r) {
		for (size_t c = 0; c < output.cols; ++c) {
			output(r, c) = target(r, c) * (*this)(r, c);
		}
	}
	return output;
}

template<typename Type>
Matrix<Type> Matrix<Type>::square() {
	Matrix output((*this));
	output = multiply_elementwise(output);
	return output;
}

template<typename Type>
Matrix<Type> Matrix<Type>::multiply_scalar(Type scalar) {
	Matrix output((*this));
	for (size_t r = 0; r < output.rows; ++r) {
		for (size_t c = 0; c < output.cols; ++c) {
			output(r, c) = scalar * (*this)(r, c);
		}
	}
	return output;
}

template<typename Type>
Matrix<Type> Matrix<Type>::add(Matrix& target) {
	assert(shape == target.shape);
	Matrix output(rows, std::get<1>(target.shape));

	for (size_t r = 0; r < output.rows; ++r) {
		for (size_t c = 0; c < output.cols; ++c) {
			output(r, c) = (*this)(r, c) + target(r, c);
		}
	}
	return output;
}

template<typename Type>
Matrix<Type> Matrix<Type>::sub(Matrix& target) {
	Matrix neg_target = -target;
	return add(neg_target);
}

template<typename Type>
Matrix<Type> Matrix<Type>::T() {
	size_t new_rows { cols }, new_cols { rows };
	Matrix transposed(new_rows, new_cols);

	for (size_t r = 0; r < new_rows; ++r) {
		for (size_t c = 0; c < new_cols; ++c) {
			transposed(r, c) = (*this)(c, r);  // swap row and col
		}
	}
	return transposed;
}

template<typename Type>
Matrix<Type> Matrix<Type>::apply_function(Type(*func)(Type)) {
	Matrix output((*this));
	for (size_t r = 0; r < rows; ++r) {
		for (size_t c = 0; c < cols; ++c) {
			output(r, c) = func((*this)(r, c));
		}
	}
	return output;
}

struct MLP {
	void printParameters();
	std::vector<Matrix<double>> weights;
	std::vector<Matrix<double>> biases;
	std::vector<Matrix<double>> layers;
	std::vector<Matrix<double>> activations;

	inline static double sigmoid(double x);
	void feedforward(Matrix<double> x);
};

void MLP::printParameters() {
	for (int i = 0; i < weights.size(); i++) {
		std::cout << "weights in layer " << i << ":" << std::endl;
		weights[i].print();
		std::cout << "biases in layer " << i << ":" << std::endl;
		biases[i].print();
	}
}

inline double MLP::sigmoid(double x) {
	return 1.0 / (1 + exp(-x));
}

/* here, passing sigmoid gives an error: argument of type
   "double (MLP::*)(double x)" is incompatible with parameter of type
"double (*)(double)" */
void MLP::feedforward(Matrix<double> x) {
	for (size_t i = 0; i < layers.size(); i++) {
		auto z = weights[i].matmul(x) + biases[i];
		activations[i] = z.apply_function(sigmoid);
	}
}

int main() {}

closed account (E093605o)
thanks, it works now.
Topic archived. No new replies allowed.