Fig. 2
From: An interpretable risk prediction model for healthcare with pattern attention
Embedding module. For each event i, we embed its happening time into vector \(v_i^t\), and its event value into vector \(v_i^e\). Then \(v_i^e\) is used to attend to the patient’s demographic information. The attention result \(v_i^j\), \(v_i^e\), \(v_i^t\) are concatenated and a fully connected layer is followed. The output vector \(q_i\) is the embedding vector for the event i