A command line utility for libraries like mlpack, weka and scikit-learn. Contains a number of utilities which can be put together with a standard unix pipeline to handle the task of training and running machine learning algorithms. The data format is meant for enabling quick trials.
All data frames are given in textual format, in which each line presents a frame for classification and needs to be prepended with a label:
NULL 1 2 3
abc 1.2 3 4
NULL 0 0 0
...
This designates three frames for classification labelled differently, each contained a three-dimensional feature vector. This can be piped directly for training:
echo "NULL 1 2 3
abc 1.2 3 4
NULL 0 0 0" | grt train RandomForests -o tt.rf
This command trains a randomforest on this dataset and puts the learned model into tt.rf. See the wiki and man-pages for more information.
grtool requires of other projects to run properly.
- ffmpeg
- python3
- a recent version of gcc
- Gesture Recognition Toolkit, please use this https://github.com/pscholl/grt
After downloading the source code install with
make PREFIX=/usr install
and install the documentation (needs pandoc) with
make PREFIX=/usr install-doc
You can run the doc-test on your machine with
make DESTDIR=test test
http://kevinboone.net/android_native.html
The scikit-learn modules (train-skl/predict-skl) are written in python3 and depend on several additional python packages. The easiest way to install them is to use pip:
pip3 install numpy scipy scikit-learn matplotlib
If you want to use the Dlib part of the tool, before installing grtool as described in the Building section:
git submodule update --init dlib
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j