机器学习

吴恩达《机器学习（2022）》

这个up做的中文翻译真的非常不错！我看视频用的是主字幕是英文，副字幕是中文，体验非常棒！

[双语人译|带测试]吴恩达2022机器学习专项课程(一）监督学习_哔哩哔哩_bilibili

[双语人译|带测试]吴恩达2022机器学习专项课程(二）：高级学习算法 Advanced Learning Algorithms_哔哩哔哩_bilibili

[双语人译|带测试]吴恩达2022机器学习专项课程(三）：无监督学习/推荐系统/强化学习_哔哩哔哩_bilibili

从2022年10月11日开学学习，在2022年11月2号学习完了所有的课程，但是代码部分没有怎么学习。学完了理论课程部分，也要着手一代码了。
这是我第一次从头到尾学习完成一个全英文的课程，其实也不算是全英文的，基本全程都在看翻译的字幕，但其实还是非常有成就感的。吴恩达老师讲的课程非常感觉非常的简单，至少入门机器学习是完全够用的了，没有讲很多数学方面的东西。当然，我也知道机器学习的理论还有非常多，仅仅学完这一门课程肯定是不够的。这门课是我探索机器学习的开始，希望我能摸索出自己的一条路。

conda install -c

jupyter book执行lab报错No module named ‘ipympl‘
【开发环境搭建】3.Anaconda安装包和channels管理

jupyter notebook添加、删除内核_IC学习者的博客-CSDN博客_jupyter notebook内核

jupyter notebook怎么切换到特定的anaconda虚拟环境_Yee_Ko的博客-CSDN博客

copy

python中的copy.deepcopy()_北木.的博客-CSDN博客_copy.deepcopy

copy.copy()是浅拷贝，只拷贝父对象，不会拷贝对象的内部的子对象。
copy.deepcopy()是深拷贝，会拷贝对象及其子对象，哪怕以后对其有改动，也不会影响其第一次的拷贝。

import copy

List1 = ['1', '2', 3, 'a', ['b', 'c']]
List2 = List1        # 将List1赋给List2
List3 = copy.copy(List1)       # 浅拷贝
List4 = copy.deepcopy(List1)   # 深拷贝

List1.append('test')     # 在List1末尾添加'test'
List1[4].append('d')     # 在List1中['b','c']的末尾添加'd'

print('List1:%s' % List1)
print('List2:%s' % List2)
print('List3:%s' % List3)
print('List4:%s' % List4)

List1:['1', '2', 3, 'a', ['b', 'c', 'd'], 'test']
List2:['1', '2', 3, 'a', ['b', 'c', 'd'], 'test']
List3:['1', '2', 3, 'a', ['b', 'c', 'd']]
List4:['1', '2', 3, 'a', ['b', 'c']]

Flynn’s taxonomy

Flynn’s taxonomy - Wikipedia
A classification of computer architectures!

• Single instruction stream, single data stream (SISD)
• Single instruction stream, multiple data streams (SIMD)
• Multiple instruction streams, single data stream (MISD)
• Multiple instruction streams, multiple data streams (MIMD)

np.c_

numpy.c_ — NumPy v1.23 Manual

sklearn

python - ModuleNotFoundError: No module named ‘sklearn’ - Stack Overflow

pip install scikit-learn

机器学习最佳Python库：Scikit-learn入门指南 - 知乎 (zhihu.com)
scikit-learn: machine learning in Python — scikit-learn 1.1.3 documentation

supervised learning

AGI: artificial general intelligence

supervised learning

1. regression: predict a number from infinitely many possible outputs.
2. classification: predict categories from small number of possible outputs.

unsupervised learning

1. clustering: group similar data points together.
2. anomaly detection: find unusual data points.
3. dimensional reduction: compress data using fewer number.

Linear Regression Model

learning algorithm
cost function

gradient descent

gradient descent algorithm
learning rate
batch gradient descent

multiple features

vectorization
feature scaling
normalization
learning curve
feature engineering

classification

logistic regression
decision boundary
cost function
gradient descent

regulation to reduce overfitting

underfit
overfit
addressing overfitting

1. collect more training examples
2. select features to include/exclude
3. regularization

Advanced algorithms

neural networks

neural networks model

1. create the model
2. loss and cost function
3. gradient descent
4. activation functions

multiclassification

softmax

optimization

Adam

CNN

machine learning diagnostic

diagnostic
evaluating
model section
training set / validation set / test set
bias / variance
baseline level of performance
learning curves

machine learning development process

iterative loop of ml development
error analysis
adding data
transfer learning

skewed datasets

error metrics: precision and recall
F1 score

decision tree

one-hot encoding
regression trees
tree ensemble

random forest algorithm

unsupervised learning

unsupervised learning

clustering

k-means

anomaly detection

recommender systems

reinforcement learning

A key point to a reinforcement learning is something called reward or reward function.

One way to think of why reinforcement learning is so powerful is you have to tell it what to do rather than how to do it.

the return in reinforcement learning

making decision: policies in reinforcement learning

markov decision process

state-action value function

Q function

bellmen equation

continuous state space