Processing order of timeseries data

[kmaehashi]

By Spark Streaming limitation, data is separated as partitions and then processed simultaneously.
As a result, training data is not ordered by time.
We should think about the way to mitigate this situation.