forest-prefetch

The original framework is provided by Sebastian Buschjäger and is extended here. Original: https://bitbucket.org/sbuschjaeger/arch-forest/src/master/

• code/ contains the actual forest and tree code synthesizer discussed in the paper
• data/ contains scripts and files for running the experiments. Each folder represents one data set used in the experiments. There are a couple of scripts for convienience. Let dataset be a dataset of choice, then
  • dataset/init.sh can be used to download and prepare this dataset. Please note, that not all data-sets can be directly downloaded via script (imdb,fact,trec). Please download those manually. The URL can be found in the init-script. Also note, that wearable-body-postures needs some manual editing of the training data, because there is a wrong line in the original file.
  • dataset/trainForest.py This trains a new RF with 25 trees on the corresponding dataset using sklearn and stores the trained model as JSON file in dataset/text/forest_25.json. Additionally, the model is exported as python pickle file in dataset/text/forsest_25.pkl
  • generateCode.py This script does the actual code generation. It receives 2 parameters. The first parameter is dataset for which code should be generated, the second one is the target architecture (arm or intel). This will generate the necessary cpp files for testing and generate a Makefile for compilation in the dataset/cpp/architechture/modelname folder.
  • compile.sh This script receives two parameters. It will compile the cpp files for the given dataset (first parameter) and target architecture (second parameter). Please make sure, that the necessary compiler is installed on your system. For intel we use g++. For arm arm-linux-gnueabihf-g++ is used.
  • run.sh This script receives two parameters. It will run the compiled cpp files for the given dataset (first parameter) and target architecture (second parameter). Results will be printed to std out. runSKLearn.sh This script receives one parameter. It receives a folder and will load the stored SKLearn model file (from the text folder) and run it on the corresponding dataset. Results will be printed to std out.
  • init_all.sh This will call the init.sh script on all folders
  • generate_all.sh This will call the generateCode.py script on all folders. It receives the target architecture as parameter
  • compile_all.sh This will call the compile.sh script on all folders.It receives the target architecture as parameter
  • run_all.sh This will call the run.sh script on all folders.It receives the target architecture as parameter

#How to

# Download the data
cd wine-quality
./init.sh

# Train the model
./trainForest.py

# Generate the code for intel
cd ..
./generatePrefetched.py wine-quality intel

# Compile it
./compile.sh wine-quality intel

# Run it
./collect.sh wine-quality intel