For LSTMTimeStep, how can I train data in the form of tensor (a, b, c)?

tensor(a, b, c):

- `a`: samples
- `b`: time steps
- `c`: features

For example tensor(3, 4, 2):


So, How can I process them as input parameters of LSTMTimeStep?

```
net.train([
{ input: [11, 12], output: [1] },
{ input: [13, 14], output: [2] },
{ input: [15, 16], output: [3] },
{ input: [17, 18], output: [4] },
])

net.train([
{ input: [21, 22], output: [5] },
{ input: [23, 24], output: [6] },
...
])

...net.train(...)
```
OR:
```
net.train([
{ input: [
[11, 12],
[13, 14],
[15, 16],
[17, 18]
],
output: [4]},
])

net.train([
{ input: [
[21, 22],
[23, 24],
...
],
output: [8]},
])

...net.train(...)
```

The question may be quite simple. Looking forward to your answer!