Suggestion: Add training_data metrics in cv return

[CliveHugo]

Environment info

Operating System: WIN10
CPU:
C++/Python/R version: python3.6

a suggestion :
Add training_data performance(metrics) in cv return for drawing validation curve (just as Xgboost.cv's return) since skleran.validation_curve is very inefficient (run n time , n =number of iteration). Don't know if anyone else has this need though.

---

Example from return of Xgboost.cv (after` pandas.from_dict):

Out[23]:
test-auc-mean test-auc-std train-auc-mean train-auc-std
0 0.847553 0.011909 0.849978 0.000601
1 0.852323 0.012572 0.856189 0.000791
2 0.856945 0.011464 0.863983 0.000770
3 0.857423 0.011400 0.869057 0.000768
4 0.857955 0.010694 0.872355 0.000579
5 0.858113 0.011444 0.874590 0.000896
6 0.857701 0.011529 0.876404 0.000870
`

[jcrosskey]

This would be a nice feature to have, it'll help monitor whether overfit occurred judging from cv.

[wjmj]

It will be a very useful feature, I hope it will be implemented as soon as possible.

[StrikerRUS]

Closed via #2089.