Fix Rank Demo Not Working

[Yunni]

Currently, the rank demo is broken. There are 2 issues with it:

• The link to download the dataset is outdated. The dataset has been moved to OneDrive. Since the data size is pretty small, I am adding the datesets directly into the repo.
• The demo script has a segment fault. The root cause is that learner -> gbm_ is loaded after learner get configured. As a result, learner -> gbm_ -> predictor is not initialized.

Also, it seems the tests for ranking is not covered very well. I'll probably create a follow-up PR to add more unit tests for ranking.

[codecov-io]

Codecov Report

Merging #2694 into master will not change coverage.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master    #2694   +/-   ##
=========================================
  Coverage     42.84%   42.84%           
  Complexity      185      185           
=========================================
  Files           150      150           
  Lines         11306    11306           
  Branches       1139     1139           
=========================================
  Hits           4844     4844           
  Misses         6137     6137           
  Partials        325      325

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e6a9063...7585a40. Read the comment docs.

[qiugen]

hi yunni. I checkout the version you fixed("The root cause is that learner -> gbm_ is loaded after learner get configured. "). while I run the bash "sh runexp.sh", program goes core dump.
runexp.sh: ? 9: 12226 ??? (??)../../xgboost mq2008.conf task=pred model_in=0004.model

[Yunni]

Interesting. It runs totally fine on my computer.
@qiugen Did you run make clean and make -j4? What's your running environment and would you mind using gdb to provide some more log?

[qiugen]

I am unfamiliar with cmake, then spent hours to find a way modify CMakelist.txt for debug info, Finally I give up, by adding '-g' in Makefile directly. I get backtrace infos as fallow:
#0 0x000000000045b4b3 in xgboost::gbm::GBTree::PredictBatch (this=0x848690, p_fmat=0x8474c0, out_preds=0x7fff0beed340, ntree_limit=0) at src/gbm/gbtree.cc:217
#1 0x000000000043425d in xgboost::LearnerImpl::PredictRaw (this=0x83f1b0, data=0x8474c0, out_preds=0x7fff0beed340, ntree_limit=0) at src/learner.cc:530
#2 0x000000000043360a in xgboost::LearnerImpl::Predict (this=0x83f1b0, data=0x8474c0, output_margin=false, out_preds=0x7fff0beed340, ntree_limit=0, pred_leaf=false, pred_contribs=false) at src/learner.cc:434
#3 0x0000000000416c12 in xgboost::CLIPredict (param=...) at src/cli_main.cc:328
#4 0x00000000004172aa in xgboost::CLIRunTask (argc=4, argv=0x7fff0beedfa8) at src/cli_main.cc:369
#5 0x0000000000417405 in main (argc=4, argv=0x7fff0beedfa8) at src/cli_main.cc:377

[Yunni]

Interesting. I got the same error when I ran it for the first time but it's fixed on both my Ubuntu and Mac after applying this PR. @qiugen Would you help me on 2 things?
(1) Would you mind providing your OS and g++ version?
(2) Can you check if the issue still exists if you swap back the 2 lines in src/cli_main.cc?

[qiugen]

（1） OS info
LSB Version: :core-3.0-amd64:core-3.0-ia32:core-3.0-noarch:graphics-3.0-amd64:graphics-3.0-ia32:graphics-3.0-noarch
Distributor ID: CentOS
Description: CentOS release 4.3 (Final)
Release: 4.3
Codename: Final
gcc version: g++ (GCC) 4.8.2
(2) after swap them back, make clean && make recompile xgboost, run sh runexp.sh the question seems to be sloved. the output pred.txt has 2874 lines in float value.