Weka中ARFF格式的属性太多

时间:2012-02-10 20:13:43

标签: weka arff

我正在使用维度超过10,000的数据集。要使用Weka,我需要将文本文件转换为ARFF格式,但由于即使在使用稀疏ARFF格式后文件太大也存在太多属性。是否有与数据类似的方法,以避免在ARFF文件的头文件中写入如此多的属性标识符。

例如:
@attribute A1 NUMERICAL
@attribute A2 NUMERICAL
           ...
           ...
@attribute A10000 NUMERICAL

1 个答案:

答案 0 :(得分:0)

我在AWK中编写了一个脚本,将以下行(在TXT文件中)格式化为ARFF

example.txt source:

Att_0 | Att_1 | Att_2 | ... | Att_n
1 | 2 | 3 | ... | 999

我的脚本(to_arff),您可以根据TXT文件中使用的分隔符更改FS值:

#!/usr/bin/awk -f
# ./<script>.awk data.txt > data.arff

BEGIN {
    FS = "|";
    # WEKA separator
    separator = ",";
}

# The first line
NR == 1 {
    # WEKA headers
        split(FILENAME, relation, ".");
        # the relation's name is the source file's name
    print "@RELATION "relation[1]"\n";
    # attributes are "numeric" by default
    # types available: numeric, <nominal> {n1, n2, ..., nN}, string and date [<date-format>]
    for (i = 1; i <= NF; i++) {
        print "@ATTRIBUTE "$i" NUMERIC";
    }
    print "\n@DATA";
}

NR > 1 {
    s = "";
    first = 1;
    for (i = 1; i <= NF; i++) {
        if (first)
            first = 0;
        else
            s = s separator;
        s = s $i;
    }
    print s;
}

输出:

@RELATION example

@ATTRIBUTE Att_0 NUMERIC
@ATTRIBUTE Att_1 NUMERIC
@ATTRIBUTE Att_2 NUMERIC
@ATTRIBUTE Att_n NUMERIC

@DATA
1,2,3,9999