当前位置:网站首页>About Tolerance Intervals
About Tolerance Intervals
2022-07-07 03:25:00 【Dreamer DBA】
It can be useful tohave an upper and lower limit on data. These bounds can be used to help identify anomalies and set expectations for what to expect.A bound on observations from a population is called a tolerance interval.
A tolerance interval is different from a prediction interval that quantifies the uncertainty for a single predicted value. It is also different from a confidence interval that quantifies the uncertainty of a population parameter such as a mean. Instead, a tolerance interval covers a proportion of the population distribution.
you will know:
- That statistical tolerance intervals provide a bounds on observations from a population.
- That a tolerance interval requires that both a coverage proportion and confidence be specified.
- That the tolerance interval for a data sample with a Gaussian distribution can be easily calculated.
1.1 Tutorial Overview
1.Bounds on Data
2. What Are Statistical Tolerance Intervals?
3. How to Calculate Tolerance Intervals
4. Tolerance Interval for Gaussian Distribution
1.2 Bounds on Data
The range of common values for data is called a tolerance interval.
1.3 What Are Statistical Tolerance Intervals?
The tolerance interval is a bound on an estimate of the proportion of data in a population.
A statistical tolerance interval [contains] a specified proportion of the units from the sampled population or process.
A tolerance interval is defined in terms of two quantities:
- Coverage: The proportion of the population covered by the interval.
- Confidence: The probabilistic confidence that the interval covers the proportion of the population.
The tolerance interval is constructed from data using two coefficients, the coverage and thetolerance coefficient. The coverage is the proportion of the population (p) that the interval is supposed to contain. The tolerance coefficient is the degree of confidence with which the interval reaches the specified coverage.
1.4 How to Calculate Tolerance Intervals
The size of a tolerance interval is proportional to the size of the data sample from the population and the variance of the population. There are two main methods for calculating tolerance intervals depending on the distribution of data: parametric and nonparametric methods.
- Parametric Tolerance Interval: Use knowledge of the population distribution in specifying both thecoverage andconfidence. Often used to refer to a Gaussian distribution.
- Nonparametric Tolerance Interval: Use rank statistics to estimate the coverage and confidence, often resulting less precision (wider intervals) given the lack of information about the distribution.
Tolerance intervals are relatively straightforward to calculate for a sample of independent observations drawn from a Gaussian distribution. We will demonstrate this calculation in the next section.
1.5 Tolerance Interval for Gaussian Distribution
We will create a sample of 100 observations drawn from a Gaussian distribution with a mean of 50 and a standard deviation of 5.
# generate dataset
from numpy.random import randn
data = 5 * randn(100) + 50
Remember that the degrees of freedom are the number of values in the calculation that can vary. Here, we have 100 observations, therefore 100 degrees of freedom. We do not know the standard deviation, therefore it must be estimated using the mean. This means our degrees of freedom will be (N - 1) or 99.
# specify degrees of freedom
n = len(data)
dof = n - 1
Next, we must specify the proportional coverage of the data.
# specify data coverage
from scipy.stats import norm
prop = 0.95
prop_inv = (1.0 - prop) / 2.0
gauss_critical = norm.ppf(prop_inv)
Next, we need to calculate the confidence of the coverage. We can do this by retrieving the critical value from the Chi-Squared distribution for the given number of degrees of freedom and desired probability. We can use the chi2.ppf() SciPy function.
# specift confidence
from scipy.stats import chi2
prob = 0.99
prop_inv = 1.0 - prob
chi_critical = chi2.ppf(prop_inv,dof)
Where dof is the number of degrees of freedom, n is the size of the data sample, gauss critical is the critical value from the Gaussian distribution, such as 1.96 for 95% coverage of the population, and chi critical is the critical value from the Chi-Squared distribution for the desired confidence and degrees of freedom.
# calculate tolerance interval
from numpy import sqrt
interval = sqrt((dof * (1 + (1/n)) * gauss_critical**2) / chi_critical)
We can tie all of this together and calculate the Gaussian tolerance interval for our data sample. The complete example is listed below.
#parametric tolerance interval
from numpy.random import seed
from numpy.random import randn
from numpy import mean
from numpy import sqrt
from scipy.stats import chi2
from scipy.stats import norm
# seed the random number generator
seed(1)
# generate dataset
data = 5 * randn(100) + 50
# specify degress of freedom
n = len(data)
dof = n - 1
# specify data coverage
prop = 0.95
prop_inv = (1.0 - prop) / 2.0
gauss_critical = norm.ppf(prop_inv)
print('Gaussian critical value: %.3f (coverage=%d%%)' %(gauss_critical,prop*100))
# specify confidence
prob = 0.99
prop_inv = 1.0 - prob
chi_critical = chi2.ppf(prop_inv, dof)
print('Chi-Squared critical value: %.3f (prob=%d%%,dof=%d)' %(chi_critical,prob*100,dof))
# tolerance
interval = sqrt((dof * (1 + (1/n)) * gauss_critical**2) / chi_critical)
print('Tolerance Interval: %.3f' % interval)
#summarize
data_mean = mean(data)
lower,upper = data_mean - interval, data_mean + interval
print('%.2f to %.2f covers %d%% of data with a confidence of %d%%' %(lower,upper,prop*100,prob*100))
Running the example first calculates and prints the relevant critical values for the Gaussian and Chi-Squared distributions. The tolerance is printed, then presented correctly.
It can also be helpful to demonstrate how the tolerance interval will decrease (become more precise) as the size of the sample is increased. The example below demonstrates this by calculating the tolerance interval for different sample sizes for the same small contrived problem.
# plot tolerance interval vs sample size
from numpy.random import seed
from numpy.random import randn
from numpy import sqrt
from scipy.stats import chi2
from scipy.stats import norm
from matplotlib import pyplot
# seed the random number generator
seed(1)
# sample sizes
seed(1)
#sample sizes
sizes = range(5,15)
for n in sizes:
# generate dataset
data = 5 * randn(n) + 50
# calculate degrees of freedom
dof = n - 1
# specify data coverage
prop = 0.95
pro_inv = (1.0 - prop) / 2.0
gauss_critical = norm.ppf(prop_inv)
# specify confidence
prob = 0.99
prop_inv = 1.0 - prob
chi_critical = chi2.ppf(prop_inv, dof)
# tolerance
tol = sqrt((dof * (1 + (1/n)) * gauss_critical**2)/ chi_critical)
# plot
pyplot.errorbar(n, 50, yerr=tol, color='blue',fmt='o')
# plot results
pyplot.show()
Running the example creates a plot showing the tolerance interval around the true population mean. We can see that the interval becomes smaller (more precise) as the sample size is increased from 5 to 15 examples.
边栏推荐
- HDU 4337 King Arthur's Knights 它输出一个哈密顿电路
- 杰理之关于 DAC 输出功率问题【篇】
- MOS transistor realizes the automatic switching circuit of main and auxiliary power supply, with "zero" voltage drop and static current of 20ua
- Not All Points Are Equal Learning Highly Efficient Point-based Detectors for 3D LiDAR Point
- An error in SQL tuning advisor ora-00600: internal error code, arguments: [kesqsmakebindvalue:obj]
- [safe office and productivity application] Shanghai daoning provides you with onlyoffice download, trial and tutorial
- 体会设计细节
- 亚像素级角点检测Opencv-cornerSubPix
- Flink task exit process and failover mechanism
- Under the tide of "going from virtual to real", Baidu AI Cloud is born from real
猜你喜欢
CVPR 2022 best paper candidate | pip: six inertial sensors realize whole body dynamic capture and force estimation
input_delay
leetcode
OC, OD, push-pull explanation of hardware
树莓派设置wifi自动连接
Shangsilicon Valley JVM Chapter 1 class loading subsystem
如何替换模型的骨干网络(backbone)
华为小米互“抄作业”
25.(arcgis api for js篇)arcgis api for js线修改线编辑(SketchViewModel)
枚举通用接口&枚举使用规范
随机推荐
CMB's written test - quantitative relationship
Matlab Error (Matrix dimensions must agree)
Flink Task退出流程与Failover机制
Function reentry, function overloading and function rewriting are understood by yourself
Household appliance industry under the "retail is king": what is the industry consensus?
About Estimation Statistics
「小样本深度学习图像识别」最新2022综述
Codeforces round 264 (Div. 2) C gargari and Bishop [violence]
[swift] learning notes (I) -- familiar with basic data types, coding styles, tuples, propositions
HMS core machine learning service creates a new "sound" state of simultaneous interpreting translation, and AI makes international exchanges smoother
MOS transistor realizes the automatic switching circuit of main and auxiliary power supply, with "zero" voltage drop and static current of 20ua
小程序能运行在自有App中,且实现直播和连麦?
[colmap] 3D reconstruction with known camera pose
上个厕所的功夫,就把定时任务的三种调度策略说得明明白白
mos管實現主副電源自動切換電路,並且“零”壓降,靜態電流20uA
HMS Core 机器学习服务打造同传翻译新“声”态,AI让国际交流更顺畅
SSL证书错误怎么办?浏览器常见SSL证书报错解决办法
2022 information security engineer examination outline
About Confidence Intervals
Flink task exit process and failover mechanism