Backend: Omikuji

The omikuji backend is a wrapper around Omikuji, an efficient implementation of a family of tree-based machine learning algorithms. It can emulate Parabel and Bonsai, two state-of-the-art algorithms for extreme multilabel classification.

The quality of results has generally been extremely good, even without tuning of hyperparameters. Training can be computationally intensive; by default it will use all available CPU cores in parallel during the training phase. Also large amounts of RAM (several GB) may be required during training, but during use the memory usage is lower.

See also the Annif-tutorial exercise about Omikuji project.

Installation

See Optional features and dependencies

Example configuration

This is a configuration that emulates Parabel. All the hyperparameters are left at their default values.

[omikuji-parabel-en]
name=Omikuji Parabel English
language=en
backend=omikuji
analyzer=snowball(english)
vocab=yso

This is a configuration that emulates Bonsai:

[omikuji-bonsai-en]
name=Omikuji Bonsai English
language=en
backend=omikuji
analyzer=snowball(english)
vocab=yso
cluster_balanced=False
cluster_k=100
max_depth=3

This is a configuration that performs AttentionXML-style collapsing of layers:

[omikuji-attention-en]
name=Omikuji Attention English
language=en
backend=omikuji
analyzer=snowball(english)
vocab=yso
cluster_balanced=False
cluster_k=2
collapse_every_n_layers=5

Backend-specific parameters

The parameters are:

Parameter	Description
limit	Maximum number of results to return
min_df	How many documents a word must appear in to be considered. Default: 1
ngram	Maximum length of word n-grams. Default: 1
cluster_balanced	Perform balanced k-means clustering instead of regular k-means clustering. Default: True
cluster_k	Number of clusters. Default: 2
max_depth	Maximum tree depth. Default: 20
collapse_every_n_layers	Number of adjacent layers to collapse. Default: 0 (disabled)

The min_df parameter controls the features (words/tokens) used to build the model. With the default setting of 1, all the words in the training set will be used, even ones that appear in only one training document. With a higher value such as 5, only those that appear in at least that many documents are included. Increasing the min_df value will decrease the size and training time of the model.

Setting the ngram parameter to 2 the vectorizer will use 2-grams as well 1-grams. This may improve the results of the model, but the model will be much larger. When using ngram>1, it probably makes sense to set min_df to something more than 1, otherwise there may be a huge number of pretty useless features.

Not all hyperparameters supported by Omikuji are currently implemented by this backend, only the ones necessary to emulate Parabel, Bonsai and the AttentionXML-like layer collapsing mode. See the omikuji README for details about the hyperparameters.

Retraining with cached training data

Preprocessing the training data can take a significant portion of the training time. If you want to experiment with different parameter settings, you can reuse the preprocessed training data by using the --cached option - see Reusing preprocessed training data. Only the analyzer, vocab and min_df settings affect the preprocessing; you can use the --cached option as long as you haven't changed these parameters.

Usage

Load a vocabulary:

annif load-vocab yso /path/to/Annif-corpora/vocab/yso-skos.ttl

Train the model:

annif train omikuji-parabel-en /path/to/Annif-corpora/training/yso-finna-en.tsv.gz

Test the model with a single document:

cat document.txt | annif suggest omikuji-parabel-en

Evaluate a directory full of files in fulltext document corpus format:

annif eval omikuji-parabel-en /path/to/documents/