Label Network Embeddings for Multi-label Classification (LNEMLC)

class skmultilearn.embedding.OpenNetworkEmbedder(graph_builder, embedding, dimension, aggregation_function, normalize_weights, param_dict=None)[source]

Bases: object

Embed the label space using a label network embedder from OpenNE

Implements an OpenNE based LNEMLC: label network embeddings for multi-label classification.

  • graph_builder (a GraphBuilderBase inherited transformer) – the graph builder to provide the adjacency matrix and weight map for the underlying graph
  • embedding (string, one of {'GraphFactorization', 'GraRep', 'HOPE', 'LaplacianEigenmaps', 'LINE', 'LLE'}) –

    the selected OpenNE embedding

    Method name string Description
    GraphFactorization Graph factorization embeddings
    GraRep Graph representations with global structural information
    HOPE High-order Proximity Preserved Embedding
    LaplacianEigenmaps Detecting communities from multiple async label propagation on the graph
    LINE Large-scale information network embedding
    LLE Locally Linear Embedding
dimension: int
the dimension of the label embedding vectors
aggregation_function: ‘add’, ‘multiply’, ‘average’ or Callable
the function used to aggregate label vectors for all labels assigned to each of the samples
normalize_weights: boolean
whether to normalize weights in the label graph by the number of samples or not
parameters passed to the embedder, don’t use the dimension and graph parameters, this class will set them at fit

If you use this classifier please cite the relevant embedding method paper and the label network embedding for multi-label classification paper:

  title={ML-KNN: A lazy learning approach to multi-label learning},
  author={Zhang, Min-Ling and Zhou, Zhi-Hua},
  journal={Pattern recognition},

Example code for using this embedder looks like this:

from skmultilearn.embedding import OpenNetworkEmbedder, EmbeddingClassifier
from sklearn.ensemble import RandomForestRegressor
from skmultilearn.adapt import MLkNN
from skmultilearn.cluster import LabelCooccurrenceGraphBuilder

graph_builder = LabelCooccurrenceGraphBuilder(weighted=True, include_self_edges=False)
openne_line_params = dict(batch_size=1000, negative_ratio=5)

clf = EmbeddingClassifier(
    OpenNetworkEmbedder(graph_builder, 'LINE', 4, 'add', True, openne_line_params),
), y_train)

predictions = clf.predict(X_test)