tmnt.inference¶

Inferencers to make predictions and analyze data using trained topic models.

Classes

`BaseInferencer`(estimator, vectorizer, device)	Base inference object for text encoding with a trained topic model.
`BowVAEInferencer`(estimator[, pre_vectorizer])
`MetricSeqVEDInferencer`(estimator, max_length)	Inferencer for sequence variational encoder-decoder models using BERT trained via Metric Learning
`SeqVEDInferencer`(estimator, max_length[, ...])	Inferencer for sequence variational encoder-decoder models using a pretrained Transformer model

class BaseInferencer(estimator, vectorizer, device)[source]¶

Base inference object for text encoding with a trained topic model.

save(model_dir)[source]¶

Save model and vectorizer to disk

Parameters:	model_dir (`str`) – Model directory to save parmaeters, config, vocabulary and vectorizer
Return type:	`None`

class BowVAEInferencer(estimator, pre_vectorizer=None)[source]¶

get_likelihood_stats(data_mat, n_samples=50)[source]¶: Get the expected elbo and its variance for input data using sampling :type data_mat: :param data_mat: Document term matrix

predict_text(txt, pred_threshold=0.5)[source]¶

Take a list of input documents/passages as strings and return document encodings (topics) and classification outputs

Parameters:	txt (`List`[`str`]) – List of input document strings pred_threshold (`float`) – Threshold to use for multilabel classification
Return type:	`Tuple`[`List`[`str`], `List`[`ndarray`], `ndarray`]
Returns:	top predicted labels, encodings, posteriors

class SeqVEDInferencer(estimator, max_length, pre_vectorizer=None, device='cpu')[source]¶

Inferencer for sequence variational encoder-decoder models using a pretrained Transformer model

class MetricSeqVEDInferencer(estimator, max_length, pre_vectorizer=None, device='cpu')[source]¶

Inferencer for sequence variational encoder-decoder models using BERT trained via Metric Learning