我试图使用scipy的entropy函数来计算KL散度。
我的p是：
array([[ 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.], [ 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [ 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [ 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], [ 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]])而q是：
array([[ 0.05242718, 0.04436347, 0.04130855, 0.04878344, 0.04310538, 0.02856853, 0.03303122, 0.02517992, 0.08525434, 0.03450324, 0.14580068, 0.1286993 , 0.28897473], [ 0.65421444, 0.11592199, 0.0642645 , 0.02989768, 0.01385762, 0.01756484, 0.01024294, 0.00891479, 0.01140301, 0.00718939, 0.00938009, 0.01070139, 0.04644726], [ 0.65984136, 0.13251236, 0.06345234, 0.02891162, 0.02429709, 0.02025307, 0.01073064, 0.01170066, 0.00678652, 0.00703361, 0.00560414, 0.00651137, 0.02236522], [ 0.32315928, 0.23900077, 0.05460232, 0.03953635, 0.02901102, 0.01294443, 0.02372061, 0.02092882, 0.01188251, 0.01377188, 0.02976672, 0.05854314, 0.14313218], [ 0.7717858 , 0.09692616, 0.03415596, 0.01713088, 0.01108141, 0.0128005 , 0.00847301, 0.01049734, 0.0052889 , 0.00514799, 0.00442508, 0.00485477, 0.01743218]], dtype=float32)当我执行以下操作时：
entropy(p[0],q[0])我得到了以下错误：
---------------------------------------------------------------------------ValueError Traceback (most recent call last)<ipython-input-201-563ea7d4decf> in <module>() 4 print('p0:',p[0]) 5 print('q0:',q[0])----> 6 entropy(p[0],q[0])/Users/freelancer/anaconda/envs/py35/lib/python3.5/site-packages/matplotlib/mlab.py in entropy(y, bins) 1570 y = np.zeros((len(x)+2,), x.dtype) 1571 y[1:-1] = x-> 1572 dif = np.diff(y) 1573 up = (dif == 1).nonzero()[0] 1574 dn = (dif == -1).nonzero()[0]/Users/freelancer/anaconda/envs/py35/lib/python3.5/site-packages/numpy/lib/function_base.py in histogram(a, bins, range, normed, weights, density) 781 if (np.diff(bins) < 0).any(): 782 raise ValueError(--> 783 'bins must increase monotonically.') 784 785 # Initialize empty histogramValueError: bins must increase monotonically.这是为什么呢？
回答：
这个例子中的数组是可以正常工作的：
import scipy as spsp.stats.entropy(p[0], q[0])从错误信息中的堆栈跟踪来看，很明显你调用的不是scipy的entropy函数，而是matplotlib的entropy函数，它们的工作方式不同。以下是相关部分：
/Users/freelancer/anaconda/envs/py35/lib/python3.5/site-packages/matplotlib/mlab.pyin entropy(y, bins)
