---

Application of OpenH in training machine learning model in hybrid

Training machine learning model (Random Forest) hybridly using OpenH

OpenH: A Programming and Execution Model for Applications on Heterogeneous Hybrid Platforms developed by heterogenous computing lab @ university college dublin

Link to paper: https://hcl.ucd.ie/system/files/OpenH.pdf

To make OpenH more portable, I reconstruct the CPU topology detection part with hwloc

---

OpenH Creators: Simon Farrelly and Ravi Reddy Manumachu

Hybrid Application Creator: Tianhao Liu

---

To laod the dataset:

1. create a dir data in <Application/forest>

2. downlaod dataset trainingdata.txt from link: https://drive.google.com/file/d/1yULAtKeRhIYNfz_aDwrIT66gtd70DDkc/view?usp=sharing

   and the dataset testingdata.txt from link: https://drive.google.com/file/d/1dtv-7rdj10L2eZL5bTIhb45JgQxnXglz/view?usp=sharing

To build and install:

1. cd <hclaffinity root directory>

2. $ source setacc.sh

3. make

4. set the env:

   export LD_LIBRARY_PATH=<dir to project root>/openhinstall/lib64:$LD_LIBRARY_PATH

5. train the machine learning model:

   cd <dir to openhinstall/Apps>
   
   ./tree_hybrid -r ../Data/trainingdata.txt -t ../Data/testingdata.txt -d 784 -c 49 -s 10000 -m 10000 -p 0
   
   ./tree_cpu -r ../Data/trainingdata.txt -t ../Data/testingdata.txt -d 784 -c 49 -s 10000 -m 10000 -p 0 -n 10
   
   ./tree_gpu -r ../Data/trainingdata.txt -t ../Data/testingdata.txt -d 784 -c 49 -s 10000 -m 10000 -p 0 -n 10

The tests and applications are installed in the openhinstall directory.

---