Junshi Guo

TINKERPOP-2487 add stdev and percentile steps

11

https://issues.apache.org/jira/browse/TINKERPOP-2487 Added two frequently used analytical steps for calculation of standard deviation and percentile value. The example usage is ``` gremlin> g.V().values('ages') ==>1 ==>2 ==>3 gremlin> g.V().values('ages').stdev() ==>0.816 gremlin> g.V().values('ages').fold().stdev(Scope.local)...

When ActivationFunc and NumHiddenNodes list sizes do not match in NN train params, the train process could still be run normally. Need to check why NN train can still work....

For mini-batch gradient update, we need to shuffle training data, so that actual training inputs for each iteration are different and random. This should benefit model performance given same iterations...