yamconv

yamconv converts a machine learning dataset from one format to another format.

Installation

yamconv is published on PyPI. You can install yamconv using pip as follows:

pip install yamconv

Alternatively, you can install it from the source code by running pip in the project directory where setup.py is located:

pip install .

Usage

yamconv.py -c converter -i input_file -o output_file -s settings -v

• -c: converter name
• -i: input file path
• -o: output file path
• -s: converter settings in JSON
• -v: verbose, to display the processing progress and information

Supported converters

The following are the supported converters:

• mlt.sqlite2fasttext: SQLite database file to fastText text file
• mlt.sqlite2csv: SQLite database file to CSV text file
• mlt.fasttext2sqlite: fastText text file to SQLite database file
• mlt.csv2sqlite: CSV text file to SQLite database file
• mlt.csv2fasttext: CSV text file to fastText text file
• mlt.sqlite2sqlite: SQLite database file to SQLite database file (with normalization)
• mlt.fasttext2fasttext: fastText text file to fastText text file (with normalization)
• mlt.csv2csv: CSV text file to CSV text file (with normalization)

Settings

Settings for converters are given in the -s option as a JSON string, e.g., '{"cache_labels": true}'.

Supported dataset formats

Multi-label text classificaiton

SQLite database

A SQLite database is used to store the classifications of texts. The database schema is as follows:

CREATE TABLE IF NOT EXISTS texts (
    id TEXT NOT NULL PRIMARY KEY,
    text TEXT NOT NULL
);
CREATE TABLE IF NOT EXISTS labels (
    label TEXT NOT NULL,
    text_id text NOT NULL,
    FOREIGN KEY (text_id) REFERENCES texts(id)
);
CREATE INDEX IF NOT EXISTS label_index ON labels (label);
CREATE INDEX IF NOT EXISTS text_id_index ON labels (text_id);

The texts table contains the text contents in the text field, and each row is uniquely identified by the id field. The labels table contains the labels in the label field. Each row has a text_id foreign key that links the label to the text in the texts table, where the text is classified with the label. In other words, each row in texts is associated with zero or more rows in labels.

fastText text file

The fastText format is a text file that contains a series of lines. Each line represents a text classified by multiple labels. A line starts with multiple labels, followed by the text content. Each label is marked with the __label__ prefix and the labels are separated by a space. The following is a fragment of an example fastText dataset file:

__label__food __label__region Many people love having dim sum in Hong Kong restaurants.
__label__region __label__plant __label__business The Netherlands is the major supplier to the European floral market.

CSV text file

The dataset is in form of a CSV (Common Separated Values) file. The first row is the header. Each of the second row and the following rows stores a single record. The CSV file can be in either of one of the following formats.

Format 1

Suppose the format of the header row is like the follwoing:

"id", "text", "region", "business", "food", "plant"

That is:

• Cell 1: id
• Cell 2: any arbitary value
• Cell n where n >= 3: the name of label n, e.g., region, business, food, plant.

Each record row looks like:

"10", "Many people love having dim sum in Hong Kong restaurants.", 1, 0, 1, 0

That is:

• Cell 1: the id string
• Cell 2: the text content
• Cell n where n >= 3: 1 or 0 representing whether the text is classified with label n or not respectively.

Format 2

Suppose the format of the header row is like the follwoing:

"text", "region", "business", "food", "plant"

That is:

• Cell 1: any arbitary value
• Cell n where n >= 2: the name of label n, e.g., region, business, food, plant.

Each record row looks like:

"Many people love having dim sum in Hong Kong restaurants.", 1, 0, 1, 0

That is:

• Cell 1: the text content
• Cell n where n >= 2: 1 or 0 representing whether the text is classified with label n or not respectively.

Profesional services

If you need any supporting resources or consultancy services from YAM AI Machinery, please find us at:

• https://www.yam.ai
• https://twitter.com/theYAMai
• https://www.linkedin.com/company/yamai
• https://www.facebook.com/theYAMai
• https://github.com/yam-ai
• https://hub.docker.com/u/yamai