| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | module athena__accuracy | ||
| 2 | !! Module containing functions to compute the accuracy of a model | ||
| 3 | use coreutils, only: real32 | ||
| 4 | implicit none | ||
| 5 | |||
| 6 | |||
| 7 | private | ||
| 8 | |||
| 9 | public :: compute_accuracy_function | ||
| 10 | public :: categorical_score | ||
| 11 | public :: mae_score, mse_score, rmse_score | ||
| 12 | public :: r2_score | ||
| 13 | |||
| 14 | |||
| 15 | abstract interface | ||
| 16 | !! Interface for the accuracy function | ||
| 17 | pure function compute_accuracy_function(predicted, expected) result(output) | ||
| 18 | !! Compute the accuracy of a model | ||
| 19 | import real32 | ||
| 20 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 21 | !! Predicted and expected values | ||
| 22 | real(real32), dimension(size(predicted,2)) :: output | ||
| 23 | !! Accuracy of the model | ||
| 24 | end function compute_accuracy_function | ||
| 25 | end interface | ||
| 26 | |||
| 27 | contains | ||
| 28 | |||
| 29 | !############################################################################### | ||
| 30 | − | pure function categorical_score(predicted, expected) result(output) | |
| 31 | !! Compute the categorical accuracy of a model | ||
| 32 | !! | ||
| 33 | !! This function is only valid for categorical/classification datasets | ||
| 34 | implicit none | ||
| 35 | |||
| 36 | !! Arguments | ||
| 37 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 38 | !! Predicted and expected values | ||
| 39 | − | real(real32), dimension(size(expected,2)) :: output | |
| 40 | !! Categorical accuracy | ||
| 41 | |||
| 42 | ! Local variables | ||
| 43 | integer :: s | ||
| 44 | !! Loop index | ||
| 45 | |||
| 46 | !! Compute the accuracy | ||
| 47 | − | do concurrent(s=1:size(expected,2)) | |
| 48 | − | if (maxloc(expected(:,s),dim=1).eq.maxloc(predicted(:,s),dim=1)) then | |
| 49 | − | output(s) = 1._real32 | |
| 50 | else | ||
| 51 | − | output(s) = 0._real32 | |
| 52 | end if | ||
| 53 | end do | ||
| 54 | |||
| 55 | − | end function categorical_score | |
| 56 | !############################################################################### | ||
| 57 | |||
| 58 | |||
| 59 | !############################################################################### | ||
| 60 | − | pure function mae_score(predicted, expected) result(output) | |
| 61 | !! Compute the mean absolute error of a model | ||
| 62 | !! | ||
| 63 | !! This function is only valid for continuous datasets | ||
| 64 | implicit none | ||
| 65 | |||
| 66 | ! Arguments | ||
| 67 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 68 | !! Predicted and expected values | ||
| 69 | − | real(real32), dimension(size(expected,2)) :: output | |
| 70 | !! Mean absolute error | ||
| 71 | |||
| 72 | ! Compute the accuracy | ||
| 73 | − | output = 1._real32 - sum(abs(expected - predicted),dim=1)/size(expected,1) | |
| 74 | |||
| 75 | − | end function mae_score | |
| 76 | !############################################################################### | ||
| 77 | |||
| 78 | |||
| 79 | !############################################################################### | ||
| 80 | − | pure function mse_score(predicted, expected) result(output) | |
| 81 | !! Compute the mean squared error of a model | ||
| 82 | !! | ||
| 83 | !! This function is only valid for continuous datasets | ||
| 84 | implicit none | ||
| 85 | |||
| 86 | ! Arguments | ||
| 87 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 88 | !! Predicted and expected values | ||
| 89 | − | real(real32), dimension(size(expected,2)) :: output | |
| 90 | !! Mean squared error | ||
| 91 | |||
| 92 | ! Compute the accuracy | ||
| 93 | − | output = 1._real32 - & | |
| 94 | − | sum((expected - predicted)**2._real32,dim=1)/size(expected,1) | |
| 95 | |||
| 96 | − | end function mse_score | |
| 97 | !############################################################################### | ||
| 98 | |||
| 99 | |||
| 100 | !############################################################################### | ||
| 101 | − | pure function rmse_score(predicted, expected) result(output) | |
| 102 | !! Compute the root mean squared error of a model | ||
| 103 | !! | ||
| 104 | !! This function is only valid for continuous datasets | ||
| 105 | implicit none | ||
| 106 | |||
| 107 | ! Arguments | ||
| 108 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 109 | !! Predicted and expected values | ||
| 110 | − | real(real32), dimension(size(expected,2)) :: output | |
| 111 | !! Root mean squared error | ||
| 112 | |||
| 113 | ! Compute the accuracy | ||
| 114 | − | output = 1._real32 - & | |
| 115 | − | sqrt(sum((expected - predicted)**2._real32,dim=1)/size(expected,1)) | |
| 116 | |||
| 117 | − | end function rmse_score | |
| 118 | !############################################################################### | ||
| 119 | |||
| 120 | |||
| 121 | !############################################################################### | ||
| 122 | − | pure function r2_score(predicted, expected) result(output) | |
| 123 | !! Compute the R^2 score of a model | ||
| 124 | !! | ||
| 125 | !! This function is only valid for continuous datasets | ||
| 126 | implicit none | ||
| 127 | |||
| 128 | ! Arguments | ||
| 129 | real(real32), dimension(:,:), intent(in) :: predicted, expected | ||
| 130 | − | real(real32), dimension(size(expected,2)) :: y_mean, rss, tss | |
| 131 | − | real(real32), dimension(size(expected,2)) :: output | |
| 132 | |||
| 133 | ! Local variables | ||
| 134 | real(real32), parameter :: epsilon = 1.E-8_real32 | ||
| 135 | !! Small value to avoid division by zero | ||
| 136 | integer :: s | ||
| 137 | !! Loop index | ||
| 138 | |||
| 139 | − | do s = 1, size(expected,2) | |
| 140 | ! compute mean of true/expected | ||
| 141 | − | y_mean(s) = sum(expected(:,s),dim=1) / size(expected,dim=1) | |
| 142 | |||
| 143 | ! compute total sum of squares | ||
| 144 | − | tss(s) = sum( ( expected(:,s) - y_mean(s) ) ** 2._real32, dim=1 ) | |
| 145 | |||
| 146 | ! compute residual sum of squares | ||
| 147 | − | rss(s) = sum( ( expected(:,s) - predicted(:,s) ) ** 2._real32, dim=1 ) | |
| 148 | |||
| 149 | ! compute accuracy (R^2 score) | ||
| 150 | − | if(abs(rss(s)).lt.epsilon)then | |
| 151 | − | output(s) = 1._real32 | |
| 152 | − | elseif(abs(tss(s)).lt.epsilon.or.rss(s)/tss(s).gt.1._real32)then | |
| 153 | − | output(s) = 0._real32 | |
| 154 | else | ||
| 155 | − | output(s) = 1._real32 - rss(s)/tss(s) | |
| 156 | end if | ||
| 157 | end do | ||
| 158 | |||
| 159 | − | end function r2_score | |
| 160 | !############################################################################### | ||
| 161 | |||
| 162 | end module athena__accuracy | ||
| 163 |