lstmCell: lstmCell
In MarkusBonsch/mxLSTM: LSTM model for regression

constructs an lstm cell. Identical to mxnet:::lstm, except for relu activation instead of tanh

1 2	lstmCell(num.hidden, indata, prev.state, param, seqidx, layeridx, dropout = 0, zoneout = 0, batchNorm = FALSE, activation = "relu")

`num.hidden`	number of hidden neurons in the cell's state
`indata`	the input to the cell
`prev.state`	the memorized state from the previous cell.
`param`	list of variables with weights and biases. must contain elements i2h.weight, i2h.bias, h2h.weight, h2h.bias
`seqidx`	sequence index. Purely bookkeeping
`layeridx`	index for the layer that the cell belongs to
`dropout`	see `mxLSTM`
`zoneout`	see `mxLSTM`
`batchNorm`	see `mxLSTM`
`activation`	activation function for update layers. "relu" or "tanh"