The keyed vectors storing the embeddings.
The list of every action the model can take.
Export the featurizer's internal parameters to be serialized along the model.
Produce an action mask according to featurizer state. (Generally, this method is reimplemented in stateful featurizers)
An array of boolean mapping every actions availability.
Turn the data returned by handleQuery into an embedding vector. This function is used to expose featurizer variables to the model optimizer for training.
Reimplementing this method is not necessary if your featurizer is not meant to be optimizable through gradient descent. In this case, just return the feature vector directly using the handleQuery method.
Let the featurizer know what action the model has taken.
Load parameters extracted from a JSON-like document.
Resets the state of the featurizer (if the stateful feature is used).
Generated using TypeDoc
Featurize queries by pooling words embedding using SWEM-concat(*).
(*): Dinghan Shen, Guoyin Wang, Wenlin Wang, Martin Renqiang Min, Qinliang Su, Yizhe Zhang, Chunyuan Li, Ricardo Henao, Lawrence Carin- 2018. Baseline Needs More Love: On Simple Word-Embedding-Based Models and Associated Pooling Mechanisms.