产生分类中的双月问题的数据集

在书”神经网络与机器学习“中，经常要用到一个数据集，就是双月数据集，作者貌似没提供生成这个数据集的代码，网上找了一个matlab版本，生成效果如下

代码如下：

function data=dbmoon(N,d,r,w)
% Usage: data=dbmoon(N,d,r,w)
% doublemoon.m - genereate the double moon data set in Haykin's book titled
% "neural networks and learning machine" third edition 2009 Pearson
% Figure 1.8 pp. 61
% The data set contains two regions A and B representing 2 classes
% each region is a half ring with radius r = 10, width = 6, one is upper
% half and the other is lower half
% d: distance between the two regions
% will generate region A centered at (0, 0) and region B is a mirror image
% of region A (w.r.t. x axis) with a (r, d) shift of origin
% N: # of samples each class, default = 1000
% d: seperation of two class, negative value means overlapping (default=1)
% r: radius (default=10), w: width of ring (default=6)
% 
% (C) 2010 by Yu Hen Hu
% Created: Sept. 3, 2010

% clear all; close all;
if nargin<4, w=6; end
if nargin<3, r=10; end
if nargin<2, d=1; end
if nargin < 1, N=1000; end

% generate region A:
% first generate a uniformly random distributed data points from (-r-w/2, 0)
% to (r+w/2, r+w/2)
N1=10*N;  % generate more points and select those meet criteria
w2=w/2; 
done=0; data=[]; tmp1=[];
while ~done, 
    tmp=[2*(r+w2)*(rand(N1,1)-0.5) (r+w2)*rand(N1,1)];
    % 3rd column of tmp is the magnitude of each data point
    tmp(:,3)=sqrt(tmp(:,1).*tmp(:,1)+tmp(:,2).*tmp(:,2)); 
    idx=find([tmp(:,3)>r-w2] & [tmp(:,3)<r+w2]);
    tmp1=[tmp1;tmp(idx,1:2)];
    if length(idx)>= N, 
        done=1;
    end
    % if not enough data point, generate more and test
end
% region A data and class label 0
% region B data is region A data flip y coordinate - d, and x coordinate +r
data=[tmp1(1:N,:) zeros(N,1);
    [tmp1(1:N,1)+r -tmp1(1:N,2)-d ones(N,1)]];

 plot(data(1:N,1),data(1:N,2),'.r',data(N+1:end,1),data(N+1:end,2),'.b');
 title(['Double moon data set, d = ' num2str(d)]),
 axis([-r-w2 2*r+w2 -r-w2-d r+w2])

save dbmoon N r w d data;

 

由于最近使用python，所以使用python翻译了一下matlab版本的代码，自己写了一个python版本

代码如下：

# Usage: data=dbmoon(N,d,r,w)
# dbmoon.py - genereate the double moon data set in Haykin's book titled
# "neural networks and learning machine" third edition 2009 Pearson
# Figure 1.8 pp. 61
# The data set contains two regions A and B representing 2 classes
# each region is a half ring with radius r = 10, width = 6, one is upper
# half and the other is lower half
# d: distance between the two regions
# will generate region A centered at (0, 0) and region B is a mirror image
# of region A (w.r.t. x axis) with a (r, d) shift of origin
# N: # of samples each class, default = 1000
# d: seperation of two class, negative value means overlapping (default=1)
# r: radius (default=10), w: width of ring (default=6)
#
# (C) 2015 by Wanqian Luo
# Created: Oct. 25, 2010

import numpy as np
def dbmoon(N=1000, d=1, r=10, w=6):
    N1 = 10*N
    w2 = w/2
    done = True
    data = np.empty(0)
    while done:
        tmp_x = 2*(r+w2)*(np.random.random([N1,1])-0.5)
        tmp_y = (r+w2)*np.random.random([N1,1])
        tmp = np.concatenate((tmp_x, tmp_y), axis=1)
        tmp_ds = np.sqrt(tmp_x*tmp_x + tmp_y*tmp_y)

        idx = np.logical_and(tmp_ds>(r-w2), tmp_ds<(r+w2))
        idx = (idx.nonzero())[0]

        if data.shape[0] == 0:
            data = tmp.take(idx,axis=0)
        else:
            data = np.concatenate((data, tmp.take(idx,axis=0)),axis=0)
        if data.shape[0] >= N:
            done = False

    db_moon = data[0:N,:]
    data_t = np.empty([N,2])
    data_t[:,0] = data[0:N,0] + r
    data_t[:,1] = -data[0:N,1] - d
    db_moon = np.concatenate((db_moon, data_t), axis=0)
    return db_moon

效果如下：

提供下代码下载，里面包含了如何调用代码的demo：http://pan.baidu.com/s/1bnsNrm3 密码：2ekd