Python 对不均衡数据进行Over sample（重抽样）

需要重采样的数据文件（Libsvm format），如heart_scale

+1 1:0.708333 2:1 3:1 4:-0.320755 5:-0.105023 6:-1 7:1 8:-0.419847 9:-1 10:-0.225806 12:1 13:-1
-1 1:0.583333 2:-1 3:0.333333 4:-0.603774 5:1 6:-1 7:1 8:0.358779 9:-1 10:-0.483871 12:-1 13:1
....

重采样后的数据保存文件（Libsvm format），这里heart_scale_balance.txt

Python code：

from sklearn.datasets import load_svmlight_file
from sklearn.datasets import dump_svmlight_file
import numpy as np
from sklearn.utils import check_random_state
from scipy.sparse import hstack,vstack

def fit_sample(X, y):
    """Resample the dataset.
    """
    label = np.unique(y)
    stats_c_ = {}
    maj_n = 0
    for i in label:
    	nk = sum(y==i)
    	stats_c_[i] = nk
    	if nk > maj_n:
    		maj_n = nk
     		maj_c_ = i

    # Keep the samples from the majority class
    X_resampled = X[y == maj_c_]
    y_resampled = y[y == maj_c_]
    # Loop over the other classes over picking at random
    for key in stats_c_.keys():

        # If this is the majority class, skip it
        if key == maj_c_:
            continue

        # Define the number of sample to create
        num_samples = int(stats_c_[maj_c_] -stats_c_[key])

        # Pick some elements at random
        random_state = check_random_state(42)
        indx = random_state.randint(low=0, high=stats_c_[key],size=num_samples)

        # Concatenate to the majority class
        X_resampled = vstack([X_resampled,X[y == key],X[y == key][indx]])
        print np.shape(y_resampled),np.shape(y[y == key]),np.shape(y[y == key][indx])
        y_resampled = list(y_resampled)+list(y[y == key])+list(y[y == key][indx])
    return X_resampled, y_resampled

X_train, y_train = load_svmlight_file("heart_scale")

# Apply the random over-sampling
X_train, y_train = fit_sample(X_train,y_train)
dump_svmlight_file(X_train, y_train,'heart_scale_balance.txt',zero_based=False)