ProbabilisticClassifierChain

Probabilistic Classifier Chains.

The Probabilistic Classifier Chains (PCC) ¹ is a Bayes-optimal method based on the Classifier Chains (CC).

Consider the concept of chaining classifiers as searching a path in a binary tree whose leaf nodes are associated with a label \(y \in Y\). While CC searches only a single path in the aforementioned binary tree, PCC looks at each of the \(2^l\) paths, where \(l\) is the number of labels. This limits the applicability of the method to data sets with a small to moderate number of labels. The authors recommend no more than about 15 labels for real-world applications.

Parameters

• model

  Type → base.Classifier

Examples

from river import linear_model
from river import metrics
from river import multioutput
from river.datasets import synth

dataset = synth.Logical(seed=42, n_tiles=100)

model = multioutput.ProbabilisticClassifierChain(
    model=linear_model.LogisticRegression()
)

metric = metrics.multioutput.MicroAverage(metrics.Jaccard())

for x, y in dataset:
   y_pred = model.predict_one(x)
   y_pred = {k: y_pred.get(k, 0) for k in y}
   metric.update(y, y_pred)
   model.learn_one(x, y)

metric

MicroAverage(Jaccard): 51.84%

Methods

learn_one

Update the model with a set of features x and the labels y.

Parameters

• x
• y
• kwargs

predict_one

Predict the labels of a set of features x.

Parameters

• x — 'dict'
• kwargs

Returns

dict[FeatureName, bool]: The predicted labels.

predict_proba_one

Predict the probability of each label appearing given dictionary of features x.

Parameters

• x
• kwargs

Returns

A dictionary that associates a probability which each label.

---

1. Cheng, W., Hüllermeier, E., & Dembczynski, K. J. (2010). Bayes optimal multilabel classification via probabilistic classifier chains. In Proceedings of the 27th international conference on machine learning (ICML-10) (pp. 279-286). ↩