Python API

Models

models.PLT(output[, tree_type, arity, …])

Probabilistic Labels Trees (PLTs) (multi-label) classifier with linear node estimators, using CPP core.

models.HSM(output[, tree_type, arity, …])

Hierarchical Softmax (multi-class) classifier with linear node estimators, using CPP core.

models.BR(output[, hash, …])

Binary Relevance (multi-label) classifier with linear estimators, using CPP core

models.OVR(output[, hash, …])

One Versus Rest (multi-class) classifier with linear estimators, using CPP core.

Datasets

datasets.download_dataset(dataset[, subset, …])

Downloads the dataset from the internet and puts it in root directory.

datasets.load_dataset(dataset[, subset, …])

Downloads the dataset from the internet and puts it in root directory.

datasets.load_libsvm_file(file)

Load data in the libsvm format into sparse CSR matrix.

datasets.to_csr_matrix(X[, shape, …])

Converts matrix-like object to Scipy csr_matrix.

Measures

measures.precision_at_k(Y_true, Y_pred[, k])

Calculate precision at 1-k places.

measures.recall_at_k(Y_true, Y_pred[, k, …])

Calculate recall at 1-k places.

measures.coverage_at_k(Y_true, Y_pred[, k])

Calculate coverage at 1-k places.

measures.dcg_at_k(Y_true, Y_pred[, k])

Calculate Discounted Cumulative Gain (DCG) at 1-k places.

measures.ndcg_at_k(Y_true, Y_pred[, k, …])

Calculate normalized Discounted Cumulative Gain (nDCG) at 1-k places.

measures.inverse_propensity(Y[, A, B])

Calculate inverse propensity as proposed in Jain et al.

measures.psprecision_at_k(Y_true, Y_pred, inv_ps)

Calculate Propensity Scored Precision (PSP) at 1-k places.

measures.psrecall_at_k(Y_true, Y_pred, inv_ps)

Calculate Propensity Scored Recall (PSR) at 1-k places.

measures.psdcg_at_k(Y_true, Y_pred, inv_ps)

Calculate Propensity Scored Discounted Cumulative Gain (PSDCG) at 1-k places.

measures.psndcg_at_k(Y_true, Y_pred, inv_ps)

Calculate Propensity Scored normalized Discounted Cumulative Gain (PSnDCG) at 1-k places.

measures.hamming_loss(Y_true, Y_pred)

Calculate unnormalized (to avoid very small numbers because of large number of labels) hamming loss - average number of misclassified labels.

measures.f1_measure(Y_true, Y_pred[, …])

Calculate F1 measure, also known as balanced F-score or F-measure.