Models

There are several models implemented currently. The best performing is at the moment the Naive Bayes algorithm.

Parameters should be under the section [model_param].

nb

SKLearn Naive Bayes model. Only works in combination with the tfidf feature extraction model. Though relatively simplistic, seems to work quite well on a wide range of datasets.

See asreview.models.NBModel

svm

SKLearn Support Vector Machine algorithm.

See asreview.models.SVMModel

rf

SKLearn Random Forest model.

See asreview.models.RFModel

logistic

SKLearn Logistic regression model.

See asreview.models.LogisticModel

nn-2-layer

Neural network consisting of 2 equal size layers. Recommended feature extraction model is doc2vec. Might crash on some systems with limited memory in combination with tfidf.

See asreview.models.NN2LayerModel

lstm-base

LSTM model that consists of an embedding layer, LSTM layer with one output, dense layer, and a single sigmoid output node. Use the embedding-lstm feature extraction method. Currently not so well optimized and slow.

See asreview.models.LSTMBaseModel

lstm-pool

LSTM model that consists of an embedding layer, LSTM layer with many outputs, max pooling layer, and a single sigmoid output node. Use the embedding-lstm feature extraction method. Currently not so well optimized and slow.

See asreview.models.LSTMPoolModel