Add hash seed feature for categorical column

fix the PSI, if there are missing columns in unit stats

…p 2. Sync column related config file like columns folder to HDFS for tracking and debugging purpose

…ing double instead of float (validate model performance at fist, then future move to float)

…ining error by weighted training size

Adopt Weight for weight optimize

# Conflicts:#src/main/java/ml/shifu/shifu/core/dtrain/wdl/WDLMaster.java#src/main/java/ml/shifu/shifu/core/dtrain/wdl/WideAndDeep.java

…d may become to 1 or 0, this case sigmoid deriveFuction is 0 without any gradient contribution, FLAT_SPOT_VALUE is 0.1 to be appended to make sure this case gradient should be considered

Add Parallel training in WDLWorker

code merge for 0.13.0 release

Fixed issue when binning field maybe NaN

Develop

…nd DOUBLE64) in NN and GBT; 2. Add new configuration to support sort or shuffle for GBT categorical variable split.

…g eval; 2.refactor the performance comparsion code - use EvalScore as input

Reduce the possibility of scam and robot scan

Add the $HOME check for shifuconfig

…ucer

…nput"This reverts commitb7dba52.

support multi-model evaluation, run shifu across cluster, and revert SE fix

norm model score name when run with -ref