update docs

tracker/README.md

@@ -0,0 +1,12 @@
+Trackers
+=====
+This folder contains tracker scripts that can be used to submit yarn jobs to different platforms,
+the example guidelines are in the script themselfs
+
+***Supported Platforms***
+* Local demo: [rabit_demo.py](rabit_demo.py)
+* MPI: [rabit_mpi.py](rabit_mpi.py)
+* Yarn (Hadoop): [rabit_yarn.py](rabit_yarn.py)
+  - It is also possible to submit via hadoop streaming with rabit_hadoop_streaming.py
+  - However, it is higly recommended to use rabit_yarn.py because this will allocate resources more precisely and fits machine learning scenarios
+

tracker/rabit_hadoop_streaming.py

@@ -1,7 +1,11 @@
 #!/usr/bin/python
 """
+Deprecated
+
 This is a script to submit rabit job using hadoop streaming.
 It will submit the rabit process as mappers of MapReduce.
+
+This script is deprecated, it is highly recommended to use rabit_yarn.py instead
 """
 import argparse
 import sys
@@ -91,6 +95,8 @@ out = out.split('\n')[0].split()
 assert out[0] == 'Hadoop', 'cannot parse hadoop version string'
 hadoop_version = out[1].split('.')
 use_yarn = int(hadoop_version[0]) >= 2
+if use_yarn:
+    warnings.warn('It is highly recommended to use rabit_yarn.py to submit jobs to yarn instead', stacklevel = 2)
 
 print 'Current Hadoop Version is %s' % out[1]
 

tracker/rabit_yarn.py

@@ -1,7 +1,7 @@
 #!/usr/bin/python
 """
-This is a script to submit rabit job using hadoop streaming.
-It will submit the rabit process as mappers of MapReduce.
+This is a script to submit rabit job via Yarn
+rabit will run as a Yarn application
 """
 import argparse
 import sys