Java Python Assignment 2 (DDL: 2024/10/20)
1. Point Estimation (15 pts)
The Poisson distribution is a useful discrete distribution which can be used to model the
number of occurrences of something per unit time. For example, in networking, packet arrival
density is often modeled with the Poisson distribution. If is Poisson distributed, i.e., its probability mass function takes the following form:
2. Source of Error: Part 1 (15 pts)
Suppose that we are given an independent and identically distributed sample of points { &}
where each point & ∼ ( , 1) is distributed according to a normal distribution with mean
and variance 1. You are going to analyze different estimators of the mean .
(a) Suppose that we use the estimator ̂= 1 for the mean of the sample, ignoring the
observed data when making our estimate. Give the bias and variance of this estimator ̂.
Explainin a sentence whether this is a good estimator in general, and give an example of
when this is a good estimator.
(b) Now suppose that we use ̂= $ as an estimator of the mean. That is, we use the first
data point in our sample to estimate the mean of the sample. Give the bias and variance
of thisestimator ̂. Explain in a sentence or two whether this is a good estimator or not.
(c) In the class you have seen the relationship between the MLE estimator and the least
squares problem. Sometimes it is useful to use the following estimate
&'$
For the mean, where the parameter > 0 is a known number. The estimator ̂ is biased,
but has lower variance than the sample mean ̅= "$ ∑& & which is an unbiased
estimator for . Give the bias and variance of the estimator ̂.
3. Source of Error: Part 2 (15 pts)
In class we discussed the fact that machine learning algorithms for function approximation
are also a kind of estimator (of the unknown target function), and that errors in function
approximation arise from three sources: bias, variance, and unavoidable error. In this part of
the question you are going to analyze error when training Bayesian classifiers. Suppose that is boolean, is real valued, ( = 1) = 1/2 and that the class conditional
distributions ( | ) are uniform distributions with ( | = 1) = [1,4] and
( | = 0) = [−4, −1]. (we use [ , ] to denote a uniform probability
distribution between and , with zero probability outside the interval [ , ]).
(a) Plot the two class conditional probability distributions ( | = 0) and ( | = 1).
(b) What is the error of the optimal classifier? Note that the optimal classifier knows ( =
1) , ( | = 0) and ( | = 1) perfectly, and applies Bayes rule to classify new
examples. Recall that the error of a classifier is the probability that it will misclassify a new
drawn at random from ( ). The error of this optimal Bayes classifier is the unavoidable
error for this learning task.
(c) Suppose instead that ( = 1) = 1/2 and that the class conditional distributions are
uniform distribution with ( | = 1) = [0,4] and ( | = 0) =
[−3,1]. What isthe unavoidable error in this case? Justify your answer.
(d) Consider again the learning task from part (a) above. Suppose we train a Gaussian Naive
Bayes (GNB) classifier using training examples for this task, where → ∞. Of course our
classifier will now (incorrectly) model ( | ) as a Gaussian distribution, so it will be
biased: it cannot even represent the correct form of ( | ) or ( | ).
Draw again the plot you created in part (a), and add to it a sketch of the learned/estimated
class conditional probability distributions the classifier will derive from the infinite training
data. Write down an expression for the error of the GNB. (hint: your expression will
involve integrals - please don't bother solving them).
(e) So far we have assumed infinite training data, so the only two sources of error are bias
and unavoidable error. Explain in one sentences how your answer to p program、c/c++,Python art (d) above would
change if the number of training examples was finite. Will the error increase or decrease?
Which of the three possible sources of error would be present in this situation?
4. Gaussian (Naïve) Bayes and Logistic Regression (15 pts)
Recall that a generative classifier estimates ( , ) = ( ) ( | ), while a discriminative
classifier directly estimates ( | ). (Note that certain discriminative classifiers are nonprobabilistic:
they directly estimate a function ∶ → instead of ( | ).) For clarity, we
highlight in bold to emphasize that it usually represents a vector of multiple attributes, i.e.,
= { $, +, . . . , %}. However, this question does not require students to derivethe answer
in vector/matrix notation.
In class we have observed an interesting relationship between a discriminative classifier
(logistic regression) and a generative classifier (Gaussian naive Bayes): the form of
( | ) derived from the assumptions of a specific class of Gaussian naive Bayes classifiers is
precisely the form used by logistic regression. The derivation can be found in the required
reading: http://www.cs.cmu.edu/~tom/mlbook/NBayesLogReg.pdf.We made the following
assumptions for Gaussian naive Bayes classifiers to model ( , ) = ( ) ( | ):
(1) is a boolean variable following a Bernouli distribution, with parameter = ( = 1)
and thus ( = 0) = 1 − .
(2) = { $, +, . . . , %}, where each attribute & is a continuous random variable. For each
& , ( &| = ) is a Gaussian distribution ( &,, &) . Note that & is the standard
deviation of the Gaussian distribution (and thus &
+ is the variance), which does not
depend on .
(3) For all ≠ , & and - are conditionally independent given . This is why this type of
classifier is called “naive”. We say this is a specific class of Gaussian naive Bayes classifiers because we have made an
assumption that the standard deviation & of ( &| = ) does not depend on the value of
. This is not a general assumption for Gaussian naive Bayes classifiers.
Let's make our Gaussian naive Bayes classifiers a little more general by removing the
assumption that the standard deviation & of ( &| = ) does not depend on . As a result,
for each &, ( &| = ) is Gaussian distribution ( &,, &,), where = 1,2, . . . , and =
0,1. Note that now the standard deviation &, of ( &| = ) depends on both the attribute
index and the value of .
Question: is the new form of ( | ) implied by this more general Gaussian naive Bayes
classifier still the form used by logistic regression? Derive the new form of ( | ) to prove
your answer.
5. Programming (40 pts)
In this lab, please submit your code according to the following guidelines:
(a) Cross-Validation: https://qffc.uic.edu.cn/home/content/index/pid/276/cid/6530.html
Please try these three approaches holdout, K-fold and leave-p-out with the data file 2.1-
Exercise.csv.
Submit ‘Exercise-handout.py’, ‘Exercise-k-fold.py’, and ‘Exercise-leave-p-out.py’
(b) Linear regression: https://qffc.uic.edu.cn/home/content/index/pid/276/cid/6541.html
Please modify linear_regression_lobf.py with the data file 2.2-Exercise.csv. For this task,
take the High column values as variables and Target column for prediction.
Submit ‘Exercise-linear_regression_lobf.py’
(c) Naïve Bayes: https://qffc.uic.edu.cn/home/content/index/pid/276/cid/6557.html
Here the dataset ‘basketball.csv’ used is for basketball games and weather conditions
where the target is if a basketball game is played in the given conditions or not, the
dataset is very small, just containing 14 rows and 5 columns.
Submit ‘Exercise-NB.py’
(d) Logistic regression: https://qffc.uic.edu.cn/home/content/index/pid/276/cid/6556.html
Use breast cancer from sklearn using following code: from sklearn.datasets import
load_breast_cancer.
Submit ‘Exercise-Logistic-Regression
相关推荐
vscode+keil嵌入式软件开发全流程
vscode+keil嵌入式软件开发全流程
讲解 program、辅导 c/c++,Python编程设计
ChampionTeam A1(bye)Team D1(School D)Team B2(School B)Team B1(School B)(bye)Team C1Team A24。
通信原理与MATLAB(八):2PSK的调制解调
通信原理和MATLAB
【自用】OMPL库windows安装教程
本文详细介绍了在Windows系统下安装OMPL库的完整流程。首先需要安装必备组件:Boost、CMake、Eigen等,并推荐使用vcpkg进行依赖管理。安装步骤包括:1)下载OMPL库源码;2)通过git克隆vcpkg;3)安装Boost及相关组件;4)将vcpkg集成到CMake;5)检查Boost安装情况;6)处理可能的路径问题;7)清理旧的CMake文件;8)使用VS2022工具进行编译;9)指定vcpkg路径进行最终编译。文中特别强调了需要确保所有依赖项版本符合要求,并提供了详细的命令行操作指南
【WRF数据准备】生成 LCZ 对应的FRC_URB2D
WRF 静态地理数据准备
华为OD机试 - 学校的位置 - 数学问题(Python/JS/C/C++ 2024 B卷 100分)
华为OD机试 2024E卷题库疯狂收录中,刷题。
program c/c++,Java
behaves.la hoflhopnothing.infectnothing.ifempty nifwall nifsame nifenemy nifenemy 4leftgo 1infectgo 1change.// worldint r;int c;
【Proteus仿真】用24C04与1602LCD设计的简易加密电子密码锁
【Proteus仿真】用24C04与1602LCD设计的简易加密电子密码锁
program c++,Python
Window:Mac:andWindow:Mac:
基于形状的模板匹配之候选点选择
基于形状的匹配匹配,顶层金字塔提取候选点
高通平台N4单路输入的720P相机预览
N4的一路720P Camera作为MIPI摄像头使用
program C++,Python
dq(d01 Solution。
手把手教你ensp仿真搭建校园网(含防火墙及无线网络)系列(一)校园网核心交换机配置
本系列博客将带你从零开始,使用 华为 eNSP(Enterprise Network Simulation Platform) 仿真搭建一个完整的校园网络,涵盖 核心层、汇聚层、接入层、防火墙安全策略、无线AC+AP组网 等实战场景。无论你是网络初学者还是希望深化实战经验的工程师,都能通过本教程掌握企业级网络的规划、配置与排错技巧。
program c++,java
Java Python This VIEW ORDER form is only used to look at existing ORDERS already in the database, and to look at the list of products ordered in each ORDER. Both your MAKE ORDER form and VIEW ORDER form should look something like FIGURE 1 (it does not need
android 蓝牙通讯demo
android 下蓝牙通讯demo,实现了发现蓝牙,使能蓝牙,通过蓝牙发送消息。
win10 跳过U8环境检测的IE WebControls.rar
win10 安装u8的时候 提示IE WebControls 不通过,这个工具可以欺骗u8的检查。 跳过U8环境检测的IE WebControls。提示导入失败后,重新检测通过。
program c++,Java
Year5 marks5 marksType5 marksthe topic.sentences.
复分析 阿尔福斯(Ahlfors) 第三版答案 英文
答案分为十章,有部分习题没有答案,全英文复分析阿尔福斯(Ahlfors) 第三版答案
Runno支持的7种编程语言:Python、Ruby、JavaScript、SQLite、C/C++、PHP完整指南
**Runno**是一个创新的沙盒化运行时环境,专门为编程语言和WASI二进制文件设计。这个强大的工具可以在浏览器、服务器或通过MCP(模型上下文协议)运行,让开发者能够安全地执行多种编程语言的代码片段。无论你是编程新手还是经验丰富的开发者,Runno都提供了一个简单而强大的平台来运行和测试代码,无需安装复杂的开发环境。 ## 🚀 Runno是什么?为什么它如此重要? Runno是一个基于W
jdk-8u144-linux-x64.tar.gz.
该资源为搭建Linux系统Java环境用,资源的后缀为tar.gz格式的压缩包
C、C++、C#、python、java编程—数据类型的使用(一)
(一)C语言 1.1数组 数组中的特定元素可以通过索引访问,第一个索引值为 0。 (1)声明 double balance[10]; (2)初始化 大括号 { } 之间的值的数目不能大于我们在数组声明时在方括号 [ ] 中指定的元素数目。 double balance[5] = {1000.0, 2.0, 3.4, 7.0, 50.0}; 如果您省略掉了数组的大小,数组的大小则为初始化时元素的个数。因此,如果: double balance[] = {1000.0, ...
C++、Python、Java多语言协同构建金融加密系统架构实战
在构建高性能、高安全性的企业级系统时,单一编程语言往往难以满足所有需求,多语言协同架构成为解决复杂工程挑战的有效方案。其核心原理在于充分发挥不同语言的特长,例如利用C++进行底层性能优化和内存安全控制,通过Java构建高并发、稳定的微服务体系,再借助Python实现快速开发和自动化运维。这种架构的技术价值在于能够综合平衡性能、开发效率和系统稳定性,尤其适用于金融科技、数据安全等对综合能力要求极高的领域。在实际应用中,通过清晰的接口抽象(如C风格API、JNI、gRPC)和统一的数据流设计,可以实现跨语言模块
PaddleHub LeViT-128 图像分类模型实战指南:安装、API 预测与 Serving 部署
本指南以 PaddleHub 图像分类模块 `levit_128_imagenet` 为核心,系统讲解 LeViT-128 混合神经网络的模型结构、基于 ImageNet-2012 预训练的权重信息,以及从安装、命令行预测、Python API 调用到在线 Serving 部署的完整实战流程。读完本文,你将掌握如何在 PaddlePaddle 生态中一键加载 LeViT-128 完成图片分类推理,
c++,c,android基础学python
大家好,我是汇源可乐。这是我的第一篇博客。我之前也学过c++,c,android{没学过Java,直接上手安卓,因为有c++基础,觉得用AS[android studio]就够了}。现在呢,因为一个比赛项目。我就转战了python. 为什么要在简书上发表个人的第一篇博客呢?因为之前在安卓编程时遇到的问题都是在简书上找-到答案的,简书的UI做的特别的好。虽然不如github优秀,但在国内,确是...
(一)c/c++、python混合编程——c/c++调用python
1.环境: win10、VS2017、Anaconda3(python3.x) 2.步骤: 2.1.在vs2017新建一个空文件test,新建文件test.c #include <Python.h> int main(int argc, char *argv[]) { Py_SetProgramName(argv[0]); Py_Initialize();...
history of program1950—2020
1.1957年 约翰·巴科斯(John Backus)创建了是全世界第一套高阶语言:FORTRAN。 2.1959年 葛丽丝·霍普(Grace Hopper)创造了现代第一个编译器A-0 系统,以及商用电脑编程语言“COBOL”,被誉为COBOL之母。 3.1964年 约翰·凯梅尼(John Kemeny)和托马斯·库尔茨(Thomas Kurtz)共同开发了BASIC程序语...
History of Program
1800年 约瑟夫·玛丽·雅卡尔(Joseph Marie Jacquard),设计出人类历史上首台可设计织布机——雅卡尔织布机,对将来发展出其他可编程机器起了重要作用。 1842年 阿达·洛夫莱斯(Ada Lovelace)为计算程序拟定“算法”,写作的第一份“程序设计流程图”,被珍视为“第一位给计算机写程序的人”。 1936年 阿兰·麦席森·图灵(Alan Mathison...
6789



被折叠的 条评论
为什么被折叠?



