forest-gis 0.0.3

A set of python modules for machine learning and data mining

Installation

Dependencies

forest-gis requires:

• Python (>= 3.6)

• NumPy (>= 1.15.0)

• SciPy (>= 0.19.1)

• joblib (>= 0.14)

For Windwos

If you already have a working installation of numpy and scipy, and you plateform is Windows 32-bit or 64-bit the easiest way to install forest-gis is using pip

pip install -U forest-gis

or conda

conda install -c conda-forge forest-gis

For linux

At present, on the pypi, we only provide wheel files supporting Python3.6, 3.7, 3.8 for Windows 32-bit, Windows 64-bit. Though the wheel files for Linux 64-bit are also provided, you may encouter problems if your Linux system has a lower version of glibc than ubantu 18.x because the wheel files was just compiled on ubantu 18.x If you get wrong when use pip to install forest-gis, you can try to install “forest-gis” from source.

For macOS

At present, install forest-gis from wheel files are not provied for macOS.

Build forest-gis from source

Before you install the forest-gis from source, you need to update cython for Windows and Linux to the newest version and then run

pip install --verbose .

For macOS, first install the macOS command line tools

brew install libomp

Set the following environment variables

export CC=/usr/bin/clang
export CXX=/usr/bin/clang++
export CPPFLAGS="$CPPFLAGS -Xpreprocessor -fopenmp"
export CFLAGS="$CFLAGS -I/usr/local/opt/libomp/include"
export CXXFLAGS="$CXXFLAGS -I/usr/local/opt/libomp/include"
export LDFLAGS="$LDFLAGS -Wl,-rpath,/usr/local/opt/libomp/lib -L/usr/local/opt/libomp/lib -lomp"

Finally, build forest-gis

pip install --verbose .

User Guide

Compute local variable importance based on decrease in node impurity

from forest.ensemble import RandomForestRegressor
rf = RandomForestRegressor(500, max_features=0.3)
rf.fit(train_x, train_y)
local_variable_importance = r_t.compute_feature_importance(X,Y,
        partition_feature = partition_feature,
                method = "lvig_based_on_impurity")

or compute local variable importance based on decrease in accuracy

from forest.ensemble import RandomForestRegressor
rf = meda.lovim(500, max_features=0.3)
rf.fit(train_x, train_y)
local_variable_importance = r_m.compute_feature_importance(X,Y,
        partition_feature = partition_feature,
                method = "lvig_based_on_accuracy")

to achieve lower computation cost, we provide a cython version based on decrease in node impurity

from forest.ensemble import RandomForestRegressor
rf = meda.lovim(500, max_features=0.3)
rf.fit(train_x, train_y)
local_variable_importance = r_m.compute_feature_importance(X,Y,
        partition_feature = partition_feature,
        method = "lvig_based_on_impurity_cython_version")