:param path: directory containing train, test, dev files :param vectors_name: name of word vectors file :param vectors_cache:

remove stop words from given sentences (questions|answers)

Random orthogonal weights Used by norm_weights(below), in which case, we are ensuring that the rows are orthogonal (i

Get the scores used to evaluate the model. Should return ([score1, score2, ..], [score1_name, score2_name, ...]). The first s

returns idf weighted overlap

returns simple overlap between document pairs

Get the map and mrr using the trec_eval utility. qids, predictions, labels should have the same length. device is not a required paramete

Random orthogonal weights Used by norm_weights(below), in which case, we are ensuring that the rows are orthogonal (i.e W = U \Sigma

Apply a feed forward network to compare o ne sentence to its soft alignment with the other. :param sentence: embedded and

Calls RetrieveSentences.getRankedPassages Parameters ---------- query : str The query to be searched in

Calls RetrieveSentences.getTermIdfJSON

returns idf weights computed over all question answer pairs in the dataset

fetches idf weights from source corpus (disks1-5+aquaint|wikipedia) index, for all the qa pairs

runs a particular combination of settings

reduce sentence terms to stemmed representations

Aggregate the representations induced from both sentences and their representations :param v1: tensor with shape (batch, time

Aggregate the representations induced from both sentences and their representations :param v1: tensor with shape (batch, tim

Compute inter-sentence attention. This is step 1 (attend) in the paper :param sent1: tensor in shape (batch, time_steps, num_units),