[英]Scipy.stats gaussian_kde to resample from conditional distribution
I am using gaussian_kde from scipy.stats to fit a joint PDF from a multivariate data on, let's say, X and Y.我正在使用来自 scipy.stats 的 gaussian_kde 来拟合来自多元数据的联合 PDF,比如 X 和 Y。
Now I want to resample from this PDF conditionally on a value of X. For example, once my X=x, generate Y from its conditional distribution.现在我想根据 X 的值有条件地从这个 PDF 重新采样。例如,一旦我的 X=x,从它的条件分布中生成 Y。
Let's use the example from the documentation here .让我们使用此处文档中的示例。
kernel.resample(1)
would generate a pair of (X,Y) over all of the distribution. kernel.resample(1)
将在所有分布上生成一对 (X,Y)。 How could I generate Y once X is, for example, 0?例如,一旦 X 为 0,我如何生成 Y?
An approach could be to create a custom continuous distribution from a pdf.一种方法可能是从 pdf 创建自定义连续分布。 The pdf can be created from the
kernel
function. pdf 可以从
kernel
function 创建。 As the pdf needs an area of 1, the kernel limited to a given x0
should be scaled by the area.由于 pdf 需要面积为 1,因此限制为给定
x0
的 kernel 应按面积缩放。
The custom distribution seems to be quite slow though.不过,自定义分发似乎很慢。 A faster solution could be to create a histogram from
ys = np.linspace(-10, 10, 1000); kernel(np.vstack([np.full_like(ys, x0), ys]))
更快的解决方案可能是从
ys = np.linspace(-10, 10, 1000); kernel(np.vstack([np.full_like(ys, x0), ys]))
创建一个直方图。 ys = np.linspace(-10, 10, 1000); kernel(np.vstack([np.full_like(ys, x0), ys]))
and use rv_histogram
. ys = np.linspace(-10, 10, 1000); kernel(np.vstack([np.full_like(ys, x0), ys]))
并使用rv_histogram
。 Still faster (but much less random) would be to use np.random.choice(..., p=...)
with p calculated from the constrained kernel.更快(但随机性要小得多)将使用
np.random.choice(..., p=...)
和 p 从受约束的 kernel 计算。
The following code starts from an adoption of the linked example code of a 2D kde.以下代码从采用 2D kde 的链接示例代码开始。
import matplotlib.pyplot as plt
from scipy import stats
import numpy as np
def measure(n):
m1 = np.random.normal(size=n)
m2 = np.random.normal(scale=0.5, size=n)
return m1 + m2, m1 - m2 ** 2
m1, m2 = measure(2000)
xmin = m1.min()
xmax = m1.max()
ymin = m2.min()
ymax = m2.max()
X, Y = np.mgrid[xmin:xmax:100j, ymin:ymax:100j]
positions = np.vstack([X.ravel(), Y.ravel()])
values = np.vstack([m1, m2])
kernel = stats.gaussian_kde(values)
Z = np.reshape(kernel(positions).T, X.shape)
x0 = 0.678
fig, (ax1, ax2) = plt.subplots(ncols=2, figsize=(10, 4))
ax1.imshow(np.rot90(Z), cmap=plt.cm.magma_r, alpha=0.4, extent=[xmin, xmax, ymin, ymax])
ax1.plot(m1, m2, 'k.', markersize=2)
ax1.axvline(x0, color='dodgerblue', ls=':')
ax1.set_xlim([xmin, xmax])
ax1.set_ylim([ymin, ymax])
# create a distribution given the kernel function limited to x=x0
class Special_distrib(stats.rv_continuous):
def _pdf(self, y, x0, area_x0):
return kernel(np.vstack([np.full_like(y, x0), y])) / area_x0
ys = np.linspace(-10, 10, 1000)
area_x0 = np.trapz(kernel(np.vstack([np.full_like(ys, x0), ys])), ys)
special_distr = Special_distrib(name="special")
vals = special_distr.rvs(x0, area_x0, size=500)
ax2.hist(vals, bins=20, color='dodgerblue')
plt.show()
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.