Metrics

Definition of XML classification and L2R metrics

Extreme Multilabel Classification

source

precision_at_k

 precision_at_k (yhat_raw, y, k=15)

Inputs: yhat_raw: activation matrix of ndarray and shape (n_samples, n_labels) y: binary ground truth matrix of type ndarray and shape (n_samples, n_labels) k: for @k metric

source

precision_at_r

 precision_at_r (yhat_raw, y)

Inputs: yhat_raw: activation matrix of ndarray and shape (n_samples, n_labels) y: binary ground truth matrix of type ndarray and shape (n_samples, n_labels)

source

recall_at_k

 recall_at_k (pred_probs, true_labels, k=15)

Learning to Rank

We want to compute a metric which measures how many orderings did the model get right:

source

batch_lbs_accuracy

 batch_lbs_accuracy (preds, xb, len=1000, resamps=10, threshold=0.5)

source

accuracy

 accuracy (xb, model)

NOTE: The following ndcg only used on a batch:

source

ndcg

 ndcg (preds, xb, k=None)

NOTE: The following ndcg_at_k only used on the entite dataset:

source

ndcg_at_k

 ndcg_at_k (dset, model, k=20)