A generalization of gcForest(or named deep Forest)

GcForest paper url： ( https://arxiv.org/abs/1702.08835 ). The official release of the gcForest code used in paper at here.

This project implements a multi-channel deep-forest based on the pylablanche's work. Thus, the project allows you to input multi-channels images.

The project implements a multi-channel deep forest so that it can process multi-channel images, such as RGB images.Besides, we use the sk-learn style with a .fit() function to train and .predict() function to predict the gcForest.
There are two sample ways to start training as follows:

The project contains a small sample data set, so you can directly run main.py for training with:

cd ./   
python main.py  

Then you can directly see the test results as follows:

Slicing Images...
sliced_imgs shape after MGS：: (2700, 972)
Training MGS Random Forests...
Adding/Training Layer, n_layer=1
Layer validation accuracy = 0.9166666666666666
Adding/Training Layer, n_layer=2
Layer validation accuracy = 0.9333333333333333
Adding/Training Layer, n_layer=3
Layer validation accuracy = 0.9333333333333333
Slicing Images...
sliced_imgs shape after MGS：: (2700, 972)
accuracy: 0.88
kappa: 0.76
              precision    recall  f1-score   support
           0       0.84      0.94      0.89       150
           1       0.93      0.82      0.87       150
    accuracy                           0.88       300
   macro avg       0.89      0.88      0.88       300
weighted avg       0.89      0.88      0.88       300
[[141   9]
 [ 27 123]]
Confusion matrix, without normalization

If you need to train gcforest with your own data：

from GCForest import *
gcf = gcForest( **kwargs ) 
gcf.fit(X_train, y_train)   # training
gcf.predict(X_test)         # inference

Using sklearn.externals.joblib save your model to disk and load it later.
save trained model :

from sklearn.externals import joblib
joblib.dump(gcf, 'trained_model.sav')

load trained model :

joblib.load('trained_model.sav')