chongshengzhang / gpuimlearn

a software for imbalance data classification based on GPU

GPU-imLearn

A software for imbalance data classification based on GPU.
It can provide high speed when you classify the multi-class imbalanced data.

This software was designed by Prof. Chongsheng ZHANG ([e-mail]), implemented by Mr. Shixin XU.

Copyright

This software is free for academic use only.

Document Description

• codes : the code work. there is a sample supported.
• data : 2 sample datasets, which are '.mat' format.
• docs : some related papers, and user manual.

Software Contents

There are 3 main parts in this work, you can see them in the codes folder.

• CPU-imLearn
  6 multi-class imbalanced classification algorithms based on CPU
• CUDA-imLearn -recommend-
  6 multi-class imbalanced classification algorithms based on GPU, base classifications are implemented by CUDA.
• PyTorch-imLearn
  6 multi-class imbalanced classification algorithms based on GPU, base classifications are implemented by Pytorch.

the 6 multi-class imbalanced classification algorithms are :

1. FocalBoost
2. DECOC
3. DOVO
4. AdaBoost.M1
5. SAMME
6. imECOC

Environment

List only the lowest.

• OS : Linux
• Python : 3
• CUDA : 9
• GCC/G++ : 6.5/6.5

Installation

• CUDA-imLearn Install before this, please install thundergbm and thundersvm.
  for more information, see ThunderGBM, and ThunderSVM.
  

>>> git clone https://github.com/inbliz/gpuimlearn.git
>>> cd gpuimlearn//src/CUDA-imLearn
>>> python3 setup.py install

• CPU-imLearn Install before this, please install thundergbm and thundersvm.
  for more information, see ThunderGBM, and ThunderSVM.
  

>>> git clone https://github.com/inbliz/gpuimlearn.git
>>> cd gpuimlearn/src/CPU-imLearn
>>> python3 setup.py install

• PyTorch-imLearn Insatll before this, please install thundergbm and thundersvm.
  for more information, see ThunderGBM, and ThunderSVM.
  

>>> git clone https://github.com/inbliz/gpuimlearn.git
>>> cd gpuimlearn/src/PyTorch-imLearn
>>> python3 setup.py install

Usage for CUDA-imLearn

They are easy to use, and you can run them quickly.
also, there is a sample in the codes folder, you can run it with the data in data folder. If this, test_sample.py and data folder shoule be in the same package.

1. DECOC Usage
   DECOC take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# DECOC
>>> from gpuimlearn import DECOC
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = DECOC.DECOC()
>>> clf.fit(X_train, y_train)
DECOC()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

2. FocalBoost Usage
   FocalBoost take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   an array y of class labels (strings or integers), of shape (n_samples),
   especially, FocalBoost supports an array imcls of the minority classes, of shape(specified by user):

# FocalBoost
>>> from gpuimlearn import FocalBoost
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = FocalBoost.FocalBoost()
>>> clf.fit(X_train, y_train)
FocalBoost()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

3. DOVO Usage
   DOVO take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# DOVO
>>> from gpuimlearn import DOVO
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = DOVO.DOVO()
>>> clf.fit(X_train, y_train)
DOVO()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

4. AdaBoost.M1 Usage
   AdaBoostM1 take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# AdaBoostM1
>>> from gpuimlearn import AdaBoostM1
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = AdaBoostM1.AdaBoostM1()
>>> clf.fit(X_train, y_train)
AdaBoostM1()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

5. SAMME Usage
   SAMME take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# SAMME
>>> from gpuimlearn import SAMME
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = SAMME.SAMME()
>>> clf.fit(X_train, y_train)
SAMME()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

6. imECOC Usage
   imECOC take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# imECOC
>>> from gpuimlearn import imECOC
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = imECOC.imECOC()
>>> clf.fit(X_train, y_train)
imECOC()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

Usage for PyTorch-imLearn & CPU-imLearn

After installation, you can use PyTorch-imLearn and CPU-imLearn like CUDA-imLearn.

Here is some samples.

1. DECOC Usage
   DECOC take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# DECOC
>>> from gpuimlearn import DECOC
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = DECOC.DECOC()
>>> clf.fit(X_train, y_train)
DECOC()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

2. FocalBoost Usage
   FocalBoost take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   an array y of class labels (strings or integers), of shape (n_samples),
   especially, FocalBoost supports an array imcls of the minority classes, of shape(specified by user):

# FocalBoost
>>> from gpuimlearn import FocalBoost
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = FocalBoost.FocalBoost()
>>> clf.fit(X_train, y_train)
FocalBoost()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])

3. DOVO Usage
   DOVO take as input two arrays: an array X of shape (n_samples, n_features) holding the training samples,
   and an array y of class labels (strings or integers), of shape (n_samples):

# DOVO
>>> from gpuimlearn import DOVO
>>> X_train = [[1, 0.5], [2, 1], [1, 1], [2, 2], [1, 2.5], [2, 4.5]]
>>> y_train = [1, 1, 2, 2, 3, 3]
>>> clf = DOVO.DOVO()
>>> clf.fit(X_train, y_train)
DOVO()

After being fitted, the model can then be used to predict new values:

>>> X_test = [[1.5, 0.5], [1.5, 1.5], [1.5, 4.5]]
>>> clf.predict(X_test)
array([1, 2, 3])