Example Machine Learning Project (SVM on the IRIS dataset)

This example project will give you a starting point for applying and/or developing machine learning algorithms for the analysis of your data. For the sake of simplicity, we will use the classical text-book example using an SVM from the scipy/numpy/sklearn stack on the IRIS dataset.

If not already done, please have a look at the example-project for the general pygradle project setup; this example assumes a working environment on your local computer.

Note that all the following commands assume that your PWD is this example project.

One-time project setup

As for the example project, you will need a setup.py in order to build the project; therefore, generate it (once) by calling:

$ ./gradlew generateSetupPy

Note that the setup.py file already exists and has been customized in order to generate an executable (see below). This required to define an entry_point for console_scripts, which now points to the main function which is responsible to train the classifier.

Dependencies

We will be using the pivy-importer to import all dependencies of our local project from pypi into a local file system path. Here, it is assumed that the pivy-importer is available on your local machine (it merely boils down to a ./gradlew assemble in that directory, and maybe copying the assembled jar to an appropriate place). For convenience, I am using the shadowJar of the pivy-importer, though the other should work as well.

For the dependencies of this sample project, you can import those using

java -jar ../../build/pivy-importer/libs/pivy-importer-0.3.36-SNAPSHOT-all.jar --repo /tmp/repo numpy:1.11.2 pandas:0.19.1 scipy:0.18.1 scikit-learn:0.18 python-dateutil:2.6.0 --replace numpy:1.6.1:numpy=1.11.2,scipy:0.9.0=scipy:0.18.1

and as a pre-requisite, make sure that development tools (python developement headers, etc.) are present such that the libraries above can be compiled.

Building the project

You are now ready to run your build, using

$ ./gradlew build

Then, after a longish build run, you should be able to call the executable in order to see the output which should be similar to this:

# ./build/deployable/bin/classify_iris
Accuracy score on the IRIS dataset using a 60/40 split: 0.966666666667

Great! You just managed to use the scipy/numpy/sklearn stack with pygradle!