README

Name

tiny-classifier

Description

On-memory text classifier command line tool, based on naive bayes.

Install

% gem install tiny-classifier

If you hope to use --tokenizer=mecab, you need to install MeCab like:

% sudo apt install mecab mecab-ipadic-utf8

This is example on Ubuntu.

Basic usage

Training:

% echo "Hello, world!"        | tc-train --labels=positive,negative positive
% echo "I'm very very happy!" | tc-train --labels=positive,negative positive
% echo "I'm so bad..."        | tc-train --labels=positive,negative negative
% echo "Oh my god!"           | tc-train --labels=positive,negative negative

The training data will be saved as tc.negative-positive.dat (tc. is the fixed prefix, .dat is the fixed suffix. The middle part is filled by given labels automatically.) in the current directory. If you hope the file to be saved in any different place, please specify --base-dir=/path/to/data/directory.

Untraining for mistakes:

% echo "I'm so bad..." | tc-untrain --labels=positive,negative positive

Testing to classify:

% echo "Happy day?" | tc-classify --labels=positive,negative
positive

If you think that the classifier has been enoughly trained, then you can generate a fixed classifier:

% tc-generate-classifier --labels=positive,negative --output-dir=/path/to/dir

Then a fixed classifier (executable Ruby script) will be generated as tc-classify-negative-positive (tc-classify- is the fixed prefix, rest is filled by given labels automatically.)

% ls /path/to/dir/
tc-classify-negative-positive
% echo "Happy day?" | /path/to/dir/tc-classify-negative-positive
positive

Command line parameters

Common

-l, --labels=LABELS (required) : A comman-separated list of labels. You should use only alphabetic characters. (Non-alphabetical characters will cause problems.)

-d, --data-dir=PATH (optional) : The path to the directory that the training data to be saved. The current directory is the default value.

-t, --tokenizer=TOKENIZER (optional) : Tokenizer for input which is not separated by whitespaces. Possible values are: only mecab.

tc-train and tc-untrain specific parameters

Both tc-train and tc-untrain require one command line argument: the label. You need to specify one of labels given via the --labels parameter.

tc-generate-classifier specific parameters

-o, --output-dir=PATH (optional) : The path to the directory that the classifier to be saved. The current directory is the default value.

Copyright

Copyright (c) 2017 YUKI "Piro" Hiroshi

License

GPLv3 or later. See LICENSE.txt for details.