matlab - 在 Matlab 中加载 LibSVM 的模型文件

Question

我需要在 Matlab 中加载 LibSVM 模型文件。Python 接口中提供了这样的功能。有人知道这个功能存在吗？

Answer 1

“要通过 MATLAB 读取数据，可以使用LIBSVM 包中的libsvmread ”

http://www.csie.ntu.edu.tw/~cjlin/libsvm/#matlab

Answer 2

我编写了一个简单的脚本来将 libsvm 中的文件加载到数据和标签中，而不使用任何 mex 文件。当然它效率不高，但它可以完成工作

file_name = 'real_sim';
fid = fopen(file_name,'r');
num_of_observations = 0;
num_of_features = 0;
num_of_nonzeros = 0;
% First pass: Extract general information
while 1
    % Read line
    line_str = fgetl(fid);
    if ~ischar(line_str)
        break;
    end    
    line_str(line_str == ':') = ' ';
    % File format: <label> <index1>:<value1> <index2>:<value2> ...
    number_list = sscanf(line_str,'%f');

    % Find number of rows, columns and nonzeros    
    feature_indices = 2:2:numel(number_list);
    temp_num = max(number_list(feature_indices));    
    if temp_num > num_of_features
        num_of_features = temp_num;
    end
    num_of_nonzeros = num_of_nonzeros + numel(feature_indices);
    num_of_observations = num_of_observations + 1;    
end
fclose(fid);

% Second pass: Read into sparse matrix
labels = zeros(num_of_observations,1);
row_indices = zeros(1,num_of_nonzeros);
col_indices = zeros(1,num_of_nonzeros);
val_list = zeros(1,num_of_nonzeros);
pos = 1;

fid = fopen(file_name,'r');
for line_num = 1 : num_of_observations
    % Read line
    line_str = fgetl(fid);
    line_str(line_str == ':') = ' ';
    % File format: <label> <index1>:<value1> <index2>:<value2> ...
    number_list = sscanf(line_str,'%f');
    % Extract data and labels
    labels(line_num) = number_list(1);
    feature_indices = 2:2:numel(number_list);
    value_indices = 3:2:numel(number_list);
    first_index = pos;
    last_index = pos + numel(feature_indices) - 1;

    row_indices(first_index:last_index) = line_num;
    col_indices(first_index:last_index) = number_list(feature_indices);
    val_list(first_index:last_index) = number_list(value_indices);

    pos = pos + numel(feature_indices);
end
fclose(fid);

data = sparse(row_indices,col_indices,val_list,num_of_observations,num_of_features);
clearvars row_indices col_indices val_list line_str 
% Convert labels to range 1-N
unique_labels = unique(labels);
num_of_labels = numel(unique_labels);
map_obj = containers.Map(unique_labels,1:num_of_labels); % Hash map
map_func = @(x) map_obj(x);
labels = arrayfun(map_func,labels);