mxLSTM

LSTM models for regression analysis with mxNet.

Features

Based on the excellent example here.

Added following features:

• takes numeric inputs instead of text (for timeseries regression)
• zoneout as described in Krueger et al. 2017 "Zoneout: Regularizing RNNs by randomly preserving hidden activations".
• batch normalization as described in T. Cooljmans et al. ILRC 2017 "Recurrent batch normalization".
• choice of different activation functions.