Python计算KL散度
import numpy as npimport scipy.statsx = [np.random.randint(1,11) for i in range(10)]print(x)print(np.sum(x))px = x/np.sum(x)#归一化print(px)y = [np.random.randint(1, 11) for i in range(10)]print(y)print(np.sum(y))py = y / np.sum(y)#归一化print(py)## scipy计算函数可以处理非归一化情况，因此这里使用# scipy.stats.entropy(x, y)或scipy.stats.entropy(px, py)均可KL = scipy.stats.entropy(x, y) print(KL)#自己编程实现kl= 0.0for i in range(10):kl += px[i] * np.log(px[i]/py[i])print(k)#TensorFlow的神经网络import sys; sys.path.append("/home/hxj/anaconda3/lib/python3.6/site-packages")import tensorflow as tfimport numpy as npx_data = np.random.rand(100).astype(np.float32)y_data = x_data*0.1+0.3print(x_data)print(y_data)Weights = tf.Variable(tf.random_uniform([1], -1.0, 1.0))biases = tf.Variable(tf.zeros([1]))y = Weights*x_data + biasesprint(y)loss = tf.reduce_mean(tf.square(y-y_data))optimizer = tf.train.GradientDescentOptimizer(0.5)train = optimizer.minimize(loss)init = tf.global_variables_initializer()sess = tf.Session()sess.run(init)for step in range(201):sess.run(train)if step % 20 == 0:print(step, sess.run(Weights), sess.run(biases))
#Python画2D图
from functools import partialimport numpyfrom matplotlib import pyplot# Define a PDFx_samples = numpy.arange(-3, 3.01, 0.01)PDF = numpy.empty(x_samples.shape)PDF[x_samples < 0] = numpy.round(x_samples[x_samples < 0] + 3.5) / 3PDF[x_samples >= 0] = 0.5 * numpy.cos(numpy.pi * x_samples[x_samples >= 0]) + 0.5PDF /= numpy.sum(PDF)# Calculate approximated CDFCDF = numpy.empty(PDF.shape)cumulated = 0for i in range(CDF.shape[0]):cumulated += PDF[i]CDF[i] = cumulated# Generate samplesgenerate = partial(numpy.interp, xp=CDF, fp=x_samples)u_rv = numpy.random.random(10000)x = generate(u_rv)# Visualizationfig, (ax0, ax1) = pyplot.subplots(ncols=2, figsize=(9, 4))ax0.plot(x_samples, PDF)ax0.axis([-3.5, 3.5, 0, numpy.max(PDF)*1.1])ax1.hist(x, 100)pyplot.show()
#Python画3D图
import matplotlib.pyplot as pltimport numpy as npfrom mpl_toolkits.mplot3d import Axes3Dnp.random.seed(42)# 采样个数500n_samples = 500dim = 3# 先生成一组3维正态分布数据，数据方向完全随机samples = np.random.multivariate_normal(np.zeros(dim),np.eye(dim),n_samples)# 通过把每个样本到原点距离和均匀分布吻合得到球体内均匀分布的样本for i in range(samples.shape[0]):r = np.power(np.random.random(), 1.0/3.0)samples[i] *= r / np.linalg.norm(samples[i])upper_samples = []lower_samples = []for x, y, z in samples:# 3x+2y-z=1作为判别平面if z > 3*x + 2*y - 1:upper_samples.append((x, y, z))else:lower_samples.append((x, y, z))fig = plt.figure('3D scatter plot')ax = fig.add_subplot(111, projection='3d')uppers = np.array(upper_samples)lowers = np.array(lower_samples)# 用不同颜色不同形状的图标表示平面上下的样本# 判别平面上半部分为红色圆点，下半部分为绿色三角ax.scatter(uppers[:, 0], uppers[:, 1], uppers[:, 2], c='r', marker='o')ax.scatter(lowers[:, 0], lowers[:, 1], lowers[:, 2], c='g', marker='^')plt.show()
