Feature types

autoBOT’s supported feature types

Feature name	Description	Code (autoBOT)
Symbolic feature types
Concept features	Features derived from ConceptNet	concept_features
Relational features - token	Features based on longer distances between tokens	relational_features_token
Relational features - char	Features based on longer distances between characters	relational_features_char
Relational features - bigram	Features based on longer distances between character pairs	relational_features_bigram
Topic features	KMeans clusters of TF-IDF into d dimensions, presences are feature values	topic_features
Keywords	Keyword-based features	keyword_features
Word-based features	Standard word TF-IDF-based features	word_features
Character-based features	Standard character TF-IDF-based features	char_features
Part-of-speech tag-based features	TF-IDF of documents, created based on POS annotations	pos_features
Sub-symbolic feature types
Neural - contextual	Features based on sentence-transformers library (XLM, multilingual)	contextual_features
Neural - dbow	doc2vec - dbow	neural_features_dbow
Neural - dm	doc2vec - dm	neural_features_dm
Document graph	Features derived as node embeddings of the corpus graph	document_graph

Representation types are defined as follows (within autoBOT); the keys of these dictionaries are the arguments under the field representation_type (see examples). Finally, note that new features are added on regular basis, hence there are more feature types available as in the original paper.

# Full stack
feature_presets['neurosymbolic'] = [
    'concept_features', 'document_graph', 'neural_features_dbow',
    'neural_features_dm', 'relational_features_token', 'topic_features',
    'keyword_features', 'relational_features_char', 'char_features',
    'word_features', 'contextual_features', 'relational_features_bigram'
]

# This one is ~language agnostic
feature_presets['neurosymbolic-lite'] = [
    'document_graph', 'neural_features_dbow', 'neural_features_dm',
    'topic_features', 'keyword_features', 'relational_features_char', 'relational_features_token','char_features', 'word_features', 'relational_features_bigram'
]

# MLJ paper versions
feature_presets['neurosymbolic-default'] = [
    'neural_features_dbow', 'neural_features_dm', 'keyword_features',
    'relational_features_char', 'char_features', 'word_features', "pos_features",
    'concept_features'
]

feature_presets['neural'] = [
    'document_graph', 'neural_features_dbow', 'neural_features_dm'
]

feature_presets['symbolic'] = [
    'concept_features', 'relational_features_token', 'topic_features',
    'keyword_features', 'relational_features_char', 'char_features',
    'word_features', 'pos_features', 'relational_features_bigram'
]

So, for example, to use the symbolic feature space only, one can simply:

autoBOTLibObj = autoBOTLib.GAlearner(train_sequences,
train_targets,
time_constraint = 1,
representation_type = "symbolic").evolve()