75bf97b575
59e63bc minor 6233050 ok 14477f9 add namenode 75a6d34 add libhdfs opts e3c76bf minmum fix 8b3c435 chg 2035799 test code 7751b2b add debug 7690313 ok bd346b4 ok faba1dc add testload 6f7783e add testload e5f0340 ok 3ed9ec8 chg e552ac4 ask for more ram in am b2505e3 only stop nm when sucess bc696c9 add queue info f3e867e add option queue 5dc843c refactor fileio cd9c81b quick fix 1e23af2 add virtual destructor to iseekstream f165ffb fix hdfs 8cc6508 allow demo to pass in env fad4d69 ok 0fd6197 fix more 7423837 fix more d25de54 add temporal solution, run_yarn_prog.py e5a9e31 final attempt ed3bee8 add command back 0774000 add hdfs to resource 9b66e7e fix hadoop 6812f14 ok 08e1c16 change hadoop prefix back to hadoop home d6b6828 Update build.sh 146e069 bugfix: logical boundary for ring buffer 19cb685 ok 4cf3c13 Merge branch 'master' of ssh://github.com/tqchen/rabit 20daddb add tracker c57dad8 add ringbased passing and batch schedule 295d8a1 update 994cb02 add sge 014c866 OK git-subtree-dir: subtree/rabit git-subtree-split: 59e63bc1354c9ff516d72d9a6468f6c431627202
Linear and Logistic Regression
- input format: LibSVM
- Local Example: run-linear.sh
- Runnig on YARN: run-yarn.sh
- You will need to have YARN
- Modify
../make/config.mkto set USE_HDFS=1 to compile with HDFS support - Run build.sh on ../../yarn on to build yarn jar file
Multi-Threading Optimization
- The code can be multi-threaded, we encourage you to use it
- Simply add
nthread=kwhere k is the number of threads you want to use
- Simply add
- If you submit with YARN
- Use
--vcoresand-memto request CPU and memory resources - Some scheduler in YARN do not honor CPU request, you can request more memory to grab working slots
- Use
- Usually multi-threading improves speed in general
- You can use less workers and assign more resources to each of worker
- This usually means less communication overhead and faster running time
Parameters
All the parameters can be set by param=value
Important Parameters
- objective [default = logistic]
- can be linear or logistic
- base_score [default = 0.5]
- global bias, recommended set to mean value of label
- reg_L1 [default = 0]
- l1 regularization co-efficient
- reg_L2 [default = 1]
- l2 regularization co-efficient
- lbfgs_stop_tol [default = 1e-5]
- relative tolerance level of loss reduction with respect to initial loss
- max_lbfgs_iter [default = 500]
- maximum number of lbfgs iterations
Optimization Related parameters
- min_lbfgs_iter [default = 5]
- minimum number of lbfgs iterations
- max_linesearch_iter [default = 100]
- maximum number of iterations in linesearch
- linesearch_c1 [default = 1e-4]
- c1 co-efficient in backoff linesearch
- linesarch_backoff [default = 0.5]
- backoff ratio in linesearch