Skip to content

Latest commit

 

History

History
106 lines (71 loc) · 3.36 KB

Readme.md

File metadata and controls

106 lines (71 loc) · 3.36 KB
Raw

Build Status

rgf_python

Machine learning Regularized Greedy Forest wrapper for Python.

Feature

Scikit learn like interface and Multi classification problem is OK.

Original rgf implementation is only available for regression and binary classification, but rgf_python is also available for Multi classification by "One-or-Rest" method.

ex.

from sklearn import datasets
from sklearn.utils.validation import check_random_state
from sklearn.cross_validation import StratifiedKFold
from rgf.rgf import RGFClassifier

iris = datasets.load_iris()
rng = check_random_state(0)
perm = rng.permutation(iris.target.size)
iris.data = iris.data[perm]
iris.target = iris.target[perm]

rgf = RGFClassifier(max_leaf=400,
                    algorithm="RGF_Sib",
                    test_interval=100,)

# cross validation
rgf_score = 0
n_folds = 3

for train_idx, test_idx in StratifiedKFold(iris.target, n_folds):
    xs_train = iris.data[train_idx]
    y_train = iris.target[train_idx]
    xs_test = iris.data[test_idx]
    y_test = iris.target[test_idx]

    rgf.fit(xs_train, y_train)
    rgf_score += rgf.score(xs_test, y_test)

rgf_score /= n_folds
print('score: {0}'.format(rgf_score))

Software Requirement

If you can't access the avove URL, alternatively, you can get RGF by downloading https://github.com/fukatani/rgf_python/releases/download/0.1.0/rgf1.2.zip. Please see README in the zip file.

Installation

git clone https://github.com/fukatani/rgf_python.git
python setup.py install

And you need to edit rgf/rgf.py

## Edit this ##################################################

#Location of the RGF executable
loc_exec = 'C:\\Users\\rf\\Documents\\python\\rgf1.2\\bin\\rgf.exe'
loc_temp = 'temp/'

## End Edit ##################################################

You need to direct actual location of rgf execution file to 'loc_exec'. 'loc_temp' is directory for placing temp file.

##Tuning hyper parameter You can tuning parameter as follows.

max_leaf: Appropriate values are data-dependent and in varied from 1000 to 10000.

test_interval: For efficiency, it must be either multiple or divisor of 100 (default of the optimization interval).

algorithm: You can select "RGF", "RGF Opt" or "RGF Sib"

loss: "LS", "Log" or "Expo".

reg_depth: Must be no smaller than 1. Meant for being used with algorithm = "RGF Opt" or "RGF Sib".

l2: Either 1, 0.1, or 0.01 often produces good results though with exponential loss (loss=Expo) and logistic loss (loss=Log) some data requires smaller values such as 1e-10 or 1e-20 Either 1, 0.1, or 0.01 often produces good results though with exponential loss (loss=Expo) and logistic loss (loss=Log) some data requires smaller values such as 1e-10 or 1e-20

sl2: Default is equal to ls. On some data, λ/100 works well.

Detail of tuning parameter is here. http://stat.rutgers.edu/home/tzhang/software/rgf/rgf1.2-guide.pdf

Using at Kaggle Kernel

Now, Kaggle Kernel supports rgh_python. Please see https://www.kaggle.com/fukatani/d/uciml/iris/classification-by-regularized-greedy-forest .

Other

Shamelessly, many part of the implementation is based on the following. Thanks! https://github.com/MLWave/RGF-sklearn