Homec4science
Diffusion Machine Learning ECMO (master) Browse

ML_ECMO/synthetic_data/data_generation/missing_value_files_creation.m
c4c453efe50dmaster

Last Change
Raw File

missing_value_files_creation.m
View Options

function missing_value_files_creation(N, N_FullNAN, Nh, N_samples, proportion_NAN, I_t, path, mode)
% Creates train/test dataset with holes
% Inputs :
% N : Number of input/output parameters
% N_FullNAN : Number of features having only NaN values
% Nh : Resolution of the grid
% proportion_NAN : array [x y] where x is the proportion of NaN
% at a given time and y the propotion of real values at the same
% time.
% I_t : Time interval
% tf : Final time
% path : path to the folder where we create the dataset
% mode : 'test' or 'train'
%
N_NotNAN = N - N_FullNAN;
% Choice between nan or 1
binary_nan = [nan 1];
for i=0:N_samples-1
% Open the files where we want to put holes
fileID1 = fopen(strcat(path, 'files_', num2str(N), '/', mode, '/target/file_', num2str(i), '.txt'),'r');
fileID2 = fopen(strcat(path, 'files_', num2str(N), '/', mode, '/input/file_', num2str(i), '.txt'),'r');
% Extract the input and output
formatSpec = '%f';
output = fscanf(fileID1, formatSpec, [N,Nh]);
input = fscanf(fileID2, formatSpec, [N,Nh]);
% Create the new files
fileID3 = fopen(strcat(path, 'files_missing_values_', num2str(N), '/', mode, '/target/file_', num2str(i), '.txt'),'w');
fileID4 = fopen(strcat(path, 'files_missing_values_', num2str(N), '/', mode, '/input/file_', num2str(i), '.txt'),'w');
input_holes = zeros(N, Nh); output_holes = zeros(N, Nh);
% Generate random unique indexes where we will put NaN values
idx_input_FullNAN = randperm(N, N_FullNAN);
idx_output_FullNAN = randperm(N, N_FullNAN);
idx_input_NotFullNAN = linspace(1, N, N);
idx_output_NotFullNAN = linspace(1, N, N);
% Remove the columns where we put NaN values, get array of shape
% N-N_NAN
idx_input_NotFullNAN(idx_input_FullNAN) = [];
idx_output_NotFullNAN(idx_output_FullNAN) = [];
for t=1:Nh
% Array full of NaN
random_binary = nan * ones(1, N_NotNAN);
% Array of one to test a condition
TRUE = ones(1, N_NotNAN);
% While the array random_binary is full of NaN we generate another
% one as we do not want NaN array.
while isnan(random_binary) == TRUE
% Create array of NaN and one with a given distribution
random_binary = randsample(binary_nan, N_NotNAN ,true, proportion_NAN);
end
% For columns where we don't want full NaN, we put some NaN at
% random using the array random_binary
output_holes(idx_output_NotFullNAN, t) = random_binary' .* output(idx_output_NotFullNAN, t);
input_holes(idx_input_NotFullNAN, t) = random_binary' .* input(idx_input_NotFullNAN, t);
end
% Interpolate the NaN values with linear interpolation
%output_holes = fillmissing(output_holes, 'linear');
%input_holes = fillmissing(input_holes, 'linear');
% Put the N_NAN columns to full NaN
output_holes(idx_output_FullNAN, :) = nan * ones(N_FullNAN, Nh);
input_holes(idx_input_FullNAN, :) = nan * ones(N_FullNAN, Nh);
% Write the resulting dataset
for j=1:N
for t=1:Nh
fprintf(fileID3,'%f ', output_holes(j, t));
fprintf(fileID4,'%f ', input_holes(j, t));
end
fprintf(fileID3, '\n');
fprintf(fileID4, '\n');
end
% Close files
fclose(fileID1);
fclose(fileID2);
fclose(fileID3);
fclose(fileID4);
end
% Plot a sample of the output and the corresponding output with holes
figure
plot(I_t, output(idx_output_NotFullNAN(1), :), I_t, output_holes(idx_output_NotFullNAN(1), :))
title(mode)
legend('sample\_output1', 'sample\_output\_holes1')
end
c4science · Help