0

我正在拟合一些数据以从对数正态分布的总和中获取参数。我正在使用石榴混合物模型。首先，生成了 3 个已知 mu 和 sigma 的种群。然后，我运行我的程序，看看石榴是否可以重现这些参数。但是，不是那么准确：

from pomegranate import *

# Generation of 3 population:
import numpy as np

s1 = np.random.lognormal(2, 0.6, size = (1000, 1))
s2 = np.random.lognormal(1.8, 0.3, size = (1000, 1))
s3 = np.random.lognormal(1.6, 0.7, size = (1000, 1))


 #universe, with all mixed populations
S =  np.concatenate((s1, s2, s3), axis=0)


#GMM

model = GeneralMixtureModel.from_samples([LogNormalDistribution], 3, S)
print(model.distributions[0].parameters[0])
print(model.distributions[1].parameters[0])
print(model.distributions[2].parameters[0])

每个种群的预期 mu 和 sigma：

s1 = (2, 0.6)
s2 = (1.8, 0.3)
s3 = (1.6, 0.7)

实际输出：

[{
    "class" : "Distribution",
    "name" : "LogNormalDistribution",
    "parameters" : [
        1.845882204858477,
        0.3306239407136521
    ],
    "frozen" : false
}]
[{
    "class" : "Distribution",
    "name" : "LogNormalDistribution",
    "parameters" : [
        2.8607217186274694,
        0.4139617889468684
    ],
    "frozen" : false
}]
[{
    "class" : "Distribution",
    "name" : "LogNormalDistribution",
    "parameters" : [
        1.6632006730938589,
        0.679604917128916
    ],
    "frozen" : false
}]

我的问题是如何使这个结果更准确？

4

0 回答 0