Classias - マニュアル

Classiasで分類器の学習を行うには，次のコマンドを用います．

$ classias-train [OPTIONS] [DATA1] [DATA2] ...

DATA1, DATA2, ...は学習データのファイルです． 学習データのファイルが指定されていないときは，標準入力から学習データを読み込みます． 学習データのファイルは，テキスト形式の他，gzip, bzip2, xzで圧縮された形式をサポートします． ファイル名の末尾が.gzで終わるときはgzip，.bz2で終わるときはbzip2，.xzで終わるときはxzユーティリティを呼び出して，学習データを自動的に解凍しながら読み込みます． 学習データのファイルが複数指定されたときは，先頭から順にグループ番号（１から開始）が割り当てられます． グループ番号は，評価用のデータセットを指定するときに使います．

コマンドラインの使用方法を見るには，-h (--help) オプションを使います．

$ classias-train -h
Classias 1.0 trainer Copyright (c) 2008,2009 Naoaki Okazaki

USAGE: classias-train [OPTIONS] [DATA1] [DATA2] ...
This utility trains a model from training data set(s).

  DATA    file(s) corresponding to data set(s) for training; if multiple N files
          are specified, this utility assigns a group number (1...N) to the
          instances in each file if no file is specified, the utility reads a
          data set from STDIN; if a file name has an extension '.gz', '.bz2',
          and '.xz', the utility uses 'gzip', 'bzip2', and 'xv' to decompress
          the input file

OPTIONS:
  -t, --type=TYPE       specify a task type (DEFAULT='multi-dense'):
      b, binary             an instance consists of a boolean class, +1 or -1,
                            and features
      m, multi-dense        an instance consists of a label and attributes;
                            features are automatically generated by combining
                            possible pairs of attributes and labels regardless
                            of the appearances in the training set
      n, multi-sparse       an instance consists of a label and attributes;
                            features are automatically generated by pairing
                            attributes and labels appearing in the training set
      c, candidate          an instance begins with a directive line '@boi'
                            followed by lines that correspond to multiple
                            candidates for the instance; a candidate line
                            consists of a class label and features; an instance
                            ends with a directive line '@eoi'
  -a, --algorithm=NAME  specify a training algorithm (DEFAULT='lbfgs.logistic')
      lbfgs.logistic        L1/L2-regularized logistic regression (LR) by L-BFGS
      averaged_perceptron   averaged perceptron
      pegasos.logistic      L2-regularized LR by Pegasos
      pegasos.hinge         L2-regularized linear L1-loss SVM by Pegasos
      truncated_gradient.logistic
                            L1-regularized LR by Truncated Gradient
      truncated_gradient.hinge
                            L1-regularized L1-loss SVM by Truncated Gradient
  -p, --set=NAME=VALUE  set the algorithm-specific parameter NAME to VALUE;
                        use '-H' or '--help-parameters' with the algorithm name
                        specified by '-a' or '--algorithm' and the task type
                        specified by '-t' or '--type' to see the list of the
                        algorithm-specific parameters
  -f, --shuffle         shuffle (reorder) instances in the data
  -b, --bias=VALUE      insert bias features with their values VALUE
  -m, --model=FILE      store the model to FILE (DEFAULT=''); if the value is
                        empty, this utility does not store the model
  -g, --split=N         split the instances into N groups; this option is
                        useful for holdout evaluation and cross validation
  -e, --holdout=M       use the M-th data for holdout evaluation and the rest
                        for training
  -x, --cross-validate  repeat holdout evaluations for #i in {1, ..., N}
                        (N-fold cross validation)
  -n, --negative=LABEL  specify a negative label for computing precision,
                        recall, and F1 scores
  -s, --token-separator=SEP assume SEP character as a token separator:
      ' ',  s, spc, space       a SPACE (' ') character (DEFAULT)
      '\t', t, tab              a TAB ('\t') character
      ',',  c, comma            a COMMA (',') character
  -c, --value-separator=SEP assume SEP character as a value separator:
      ':',  c, colon            a COLON (':') character (DEFAULT)
      '=',  e, equal            a EQUAL ('=') character
      '|',  b, bar              a BAR ('|') character
  -h, --help            show this help message and exit
  -H, --help-parameters show the help message of algorithm-specific parameters;
                        specify an algorithm with '-a' or '--algorithm' option,
                        and specify a task type with '-t' or '--task' option

以下，各オプションを説明します．

学習で獲得した分類器を適用するには，次のコマンドを用います．

$ classias-tag [OPTIONS]

分類を行うデータは標準入力から与えられるものと仮定します． 分類のタスク・タイプは，-mオプションで指定したモデルから自動的に決定されます． 分類を行うデータの形式は，学習に用いるデータ形式と同一で，ラベルのフィールドを省略することはできません（事例のラベルは空文字列にしてください）． 事例の正解ラベルが分かっていて，分類器の評価に利用したいときは，入力データに事例のラベルをセットしてください． 分類結果は標準出力に出力されます．

コマンドラインの使用方法を見るには，-h (--help) オプションを使います．

$ classias-tag  -h
USAGE: classias-tag [OPTIONS]
This utility tags labels for a data set read from STDIN.

OPTIONS:
  -m, --model=FILE      load the model from FILE
  -t, --test            evaluate the tagging performance on the labeled data
  -n, --negative=LABEL  assume LABEL to be a negative label
  -w, --score           output scores for the labels
  -p, --probability     output probabilities for the labels
  -r, --reference       output reference labels as well as predicted labels
  -k, --comment         output comment lines in the tagging output
  -a, --all             output all candidate labels in the tagging output
  -f, --false           output false instances only
  -q, --quiet           suppress tagging results from the output
  -s, --token-separator=SEP assume SEP character as a token separator:
      ' ',  s, spc, space       a SPACE (' ') character (DEFAULT)
      '\t', t, tab              a TAB ('\t') character
      ',',  c, comma            a COMMA (',') character
  -c, --value-separator=SEP assume SEP character as a value separator:
      ':',  c, colon            a COLON (':') character (DEFAULT)
      '=',  e, equal            a EQUAL ('=') character
      '|',  b, bar              a BAR ('|') character
  -v, --version         show the version and copyright information
  -h, --help            show this help message and exit

以下，各オプションを説明します．