Multilabel classification
Multilabel classification is a classification task in which each input may be assigned zero or more labels simultaneously.
Natural Language Processing in Action, Second Edition
Multi-label classification assigns zero or more independent labels to each input example.
- How it is used
- Assigns overlapping toxicity or intent tags to text when forcing one mutually exclusive class would lose useful information.
- How deep it goes
- Defines independent tags and multi-hot targets, then demonstrates dataset preparation, pretrained-model fine-tuning, and evaluation without deriving model internals.
- Examples
- Uses ambiguous surnames, six-label toxic-comment rows, and scored math-tutor intents to show why several tags may apply.
- What you can do after
- Applies simpletransformers’ MultiLabelClassificationModel to fine-tune and evaluate a BERT toxic-comment tagger.
Read how Natural Language Processing in Action, Second Edition explains multilabel classification →
Deep Learning with Python, Third Edition
A classification task in which each example may belong to multiple classes at the same time, even though there are multiple possible classes overall.
- How it is used
- Frames classification tasks where each sample may receive several labels, guiding the choice of output layer, loss, and targets.
- How deep it goes
- Defines the task and gives a Keras output-layer and loss recipe, without a worked multilabel training run.
- Examples
- Uses a cat-and-dog image, a photo search task, and a hypothetical multilabel version of newswire topic classification.
- What you can do after
- Apply it via Keras by configuring one sigmoid unit per class, binary_crossentropy loss, and k-hot targets.
Read how Deep Learning with Python, Third Edition explains multilabel classification →
Deep Learning with R, Third Edition
Multilabel multiclass classification is a classification task in which each input may belong to multiple categories drawn from a set of more than two classes.
- How it is used
- Frames prediction tasks where one input may receive several category labels, especially image tagging and topic assignment.
- How deep it goes
- Defines multilabel classification and contrasts it with binary and single-label multiclass classification, without detailing model training.
- Examples
- Uses cat-and-dog image tagging, Reuters newswire topics, and Google Photos search across more than 20,000 image classes.
- What you can do after
- Evaluate/understand whether a classification problem permits multiple labels per input rather than mutually exclusive categories.
Read how Deep Learning with R, Third Edition explains multilabel classification →
Related concepts
- classifierkind of · A classifier is a predictive computational model that assigns inputs to one or more discrete classes or categories.
- single-label categorical classificationcontrasts with · A classification setup in which each input is assigned no more than one label from a set of classes.
- multi-hot tensoruses · A multi-hot tensor is a binary tensor in which multiple positions can be active simultaneously to represent selected labels or attributes.
- multiclass classificationkind of · Multiclass classification is a supervised prediction task that assigns each input to exactly one of more than two possible classes.
- classificationkind of · Classification is the task of assigning an input or observation to one or more categories from a defined set of labels.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.