正文

43 - What is machine learning anyway

Machine Learning 机器学习
- Supervised Learning 监督学习
  - Classification problem 分类问题
    - Support Vector Machines 支持向量机
    - Discriminant Analysis 判别分析法
    - Naive Bayes 朴素贝叶斯
    - Nearest Neighbor 邻近算法
  - Regression 回归
    - Linear Regression, GLM 线性回归，广义线性模型
    - SVR, GPR 支持向量回归，高斯过程回归
    - Ensemble Methods 集成学习算法
    - Decision Trees 决策树
    - Neural Networks 神经网络
- Unsupervised Learning 无监督学习
  - Clustering 聚类
    - K-Means, K-Medoids, Fuzzy C-Means
    - Hierarchical 层次聚类
    - Gaussian Mixture 高斯混合模型
    - Neural Networks 神经网络
    - Hidden Markov Model 隐马尔科夫模型

Machine Learning VS Deep Learning

在数据量大的情况下使用深度学习效果更好。

44 - What is linear regression

介绍了下线性回归以及损失函数。

45 - Linear regression using Sci-Kit Learn in Python

python

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn import linear_model
 
df = pd.read_csv('data/cells.csv')
df

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

python

plt.xlabel('time')
plt.ylabel('cells')
plt.scatter(df.time, df.cells, color='red', marker='+')

<matplotlib.collections.PathCollection at 0x2c7bd4eb0d0>

x independent (time)
y dependent - we are predicting Y

python

x_df = df[['time']]
x_df

	time
0	0.0
1	0.5
2	1.0
3	1.5
4	2.0
5	2.5
6	3.0
7	3.5
8	4.0
9	4.5
10	5.0

python

x_df.dtypes

time    float64
dtype: object

python

y_df = df.cells

创建并训练模型

python

reg = linear_model.LinearRegression()  # Create an instance of the model
reg.fit(x_df, y_df)  # Training the model (fitting a line)

预测模型

python

# Predict
reg.predict([[2.3]])

C:\Users\gzjzx\anaconda3\lib\site-packages\sklearn\base.py:450: UserWarning: X does not have valid feature names, but LinearRegression was fitted with feature names
  warnings.warn(



array([257.61090909])

评分

python

reg.score(x_df, y_df)

0.9784252641866715

Y = mx + C

python

c = reg.intercept_
m = reg.coef_
2.3 * m + c

array([257.61090909])

预测多个数据

python

cells_predict_df = pd.read_csv('data/cells_predict.csv')
cells_predict_df.head()

	time
0	0.1
1	0.2
2	0.3
3	0.4
4	0.5

python

predicted_cells = reg.predict(cells_predict_df)
predicted_cells

array([212.33090909, 214.38909091, 216.44727273, 218.50545455,
       220.56363636, 222.62181818, 224.68      , 226.73818182,
       228.79636364, 230.85454545, 232.91272727, 234.97090909,
       237.02909091, 239.08727273, 241.14545455, 243.20363636,
       245.26181818, 247.32      , 249.37818182, 251.43636364,
       253.49454545, 255.55272727, 257.61090909, 259.66909091,
       261.72727273, 263.78545455, 265.84363636, 267.90181818,
       269.96      , 272.01818182, 274.07636364, 276.13454545,
       278.19272727, 280.25090909, 282.30909091, 284.36727273,
       286.42545455, 288.48363636, 290.54181818, 292.6       ])

将预测数据合并并保存

python

cells_predict_df['cells'] = predicted_cells
cells_predict_df.head()

	time	cells
0	0.1	212.330909
1	0.2	214.389091
2	0.3	216.447273
3	0.4	218.505455
4	0.5	220.563636

python

cells_predict_df.to_csv('predicted_cells.csv')

46 - Splitting data into training and testing sets for machine learning

python

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn import linear_model

python

df = pd.read_csv('data/cells.csv')
df

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

python

x_df = df.drop('cells', axis='columns')
y_df = df.cells

分割训练集(train)和测试集(test)

python

from sklearn.model_selection import train_test_split
 
X_train, X_test, y_train, y_test = train_test_split(x_df, y_df, test_size=0.4, random_state=10)

python

X_train

	time
10	5.0
3	1.5
1	0.5
0	0.0
4	2.0
9	4.5

python

reg = linear_model.LinearRegression()
reg.fit(X_train, y_train)
 
prediction_test = reg.predict(X_test)
prediction_test

array([229.66081871, 270.73684211, 291.2748538 , 260.46783626,
       281.00584795])

计算均方误差

python

print('Mean sq. error between y_test and predicted =', np.mean(prediction_test - y_test) ** 2)

Mean sq. error between y_test and predicted = 7.677112273861912

计算残差

python

plt.scatter(prediction_test, prediction_test - y_test)
plt.hlines(y=0, xmin=200, xmax=310)  # 画一条水平线

<matplotlib.collections.LineCollection at 0x26a2470d640>

47 - Multiple Linear Regression with SciKit-Learn in Python

python

import pandas as pd
 
df = pd.read_excel('data/images_analyzed.xlsx')
df.head()

	User	Time	Coffee	Age	Images_Analyzed
0	1	8	0	23	20
1	1	13	0	23	14
2	1	17	0	23	18
3	1	22	0	23	15
4	1	8	2	23	22

python

import seaborn as sns
 
sns.lmplot(x='Time', y='Images_Analyzed', data=df, hue='Age')

<seaborn.axisgrid.FacetGrid at 0x238adc47910>

python

import numpy as np
from sklearn import linear_model
 
reg = linear_model.LinearRegression()
reg.fit(df[['Time', 'Coffee', 'Age']], df.Images_Analyzed)
 
reg.coef_, reg.intercept_

(array([-0.35642282, -0.3475    , -0.04279945]), 25.189636192124166)

python

reg.predict([[13, 2, 23]])

C:\Users\gzjzx\anaconda3\lib\site-packages\sklearn\base.py:450: UserWarning: X does not have valid feature names, but LinearRegression was fitted with feature names
  warnings.warn(



array([18.8767522])

48 - What is logistic regression

虽然是 Regression，不过主要用于 Classfication （二分类问题）。

逻辑回归（Logistic Regression）（一） - 知乎 (zhihu.com)

49 - Logistic Regression using scikit-learn in Python

STEP1: DATA READING AND UNDERSTANDING

python

import pandas as pd
from matplotlib import pyplot as plt
 
df = pd.read_csv('data/images_analyzed_productivity1.csv')
df.head()

	User	Time	Coffee	Age	Images_Analyzed	Productivity
0	1	8	0	23	20	Good
1	1	13	0	23	14	Bad
2	1	17	0	23	18	Good
3	1	22	0	23	15	Bad
4	1	8	2	23	22	Good

python

plt.scatter(df.Time, df.Productivity, marker='+', color='red')

<matplotlib.collections.PathCollection at 0x206c9140bb0>

python

sizes = df['Productivity'].value_counts(sort=1)
plt.pie(sizes, autopct='%1.1f%%')

([<matplotlib.patches.Wedge at 0x206cc3afb80>,
  <matplotlib.patches.Wedge at 0x206cc3b9310>],
 [Text(-0.08630492316306847, 1.096609073570804, ''),
  Text(0.08630482049111692, -1.0966090816512493, '')],
 [Text(-0.04707541263440097, 0.598150403765893, '52.5%'),
  Text(0.04707535663151831, -0.5981504081734086, '47.5%')])

STEP2: DROP IRRLEVANT DATA 丢弃无关数据

python

df.drop(['Images_Analyzed'], axis=1, inplace=True)
df.drop(['User'], axis=1, inplace=True)
df.head()

	Time	Coffee	Age	Productivity
0	8	0	23	Good
1	13	0	23	Bad
2	17	0	23	Good
3	22	0	23	Bad
4	8	2	23	Good

STEP3: DEAL WITH MISSING VALUES 处理缺失数据

python

df = df.dropna()

STEP4: CONVERT NON-NUMERIC TO NUMERIC

将 Good 和 Bad 替换成计算机便于理解的数字类型

python

df.Productivity[df.Productivity == 'Good'] = 1
df.Productivity[df.Productivity == 'Bad'] = 2
df.head()

	Time	Coffee	Age	Productivity
0	8	0	23	1
1	13	0	23	2
2	17	0	23	1
3	22	0	23	2
4	8	2	23	1

STEP 5: PREPARE THE DATA(define indep/dep variables)

python

Y = df['Productivity'].values
Y = Y.astype('int')
Y

array([1, 2, 1, 2, 1, 2, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 1, 2, 1, 2, 1, 2,
       1, 2, 1, 2, 1, 2, 1, 2, 2, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 1,
       1, 2, 2, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 2, 2, 1, 2,
       1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 1, 2, 2, 2])

python

X = df.drop(labels=['Productivity'], axis=1)
X.head()

	Time	Coffee	Age
0	8	0	23
1	13	0	23
2	17	0	23
3	22	0	23
4	8	2	23

STEP 6: SPLIT DATA

python

from sklearn.model_selection import train_test_split
 
X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.1, random_state=20)
X_train.head()

	Time	Coffee	Age
42	18	4	31
5	13	2	23
54	17	2	45
12	8	6	23
78	17	6	52

STEP 7: DEFINE THE MODEL

python

from sklearn.linear_model import LogisticRegression
 
model = LogisticRegression()
model.fit(X_train, y_train)

STEP 8: TESTING THE MODEL

python

prediction_test = model.predict(X_test)
prediction_test

array([2, 2, 2, 1, 1, 2, 1, 2])

STEP 9: VERIFY THE ACCURACY 验证准确性

python

from sklearn import metrics
 
print('Accuracy =', metrics.accuracy_score(y_test, prediction_test))

Accuracy = 0.75

STEP 10: WEIGHTS

python

model.coef_

array([[0.18788991, 0.19204588, 0.0200644 ]])

python

weights = pd.Series(model.coef_[0], index=X.columns.values)
weights

Time      0.187890
Coffee    0.192046
Age       0.020064
dtype: float64

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309

	time	cells
0	0.0	205
1	0.5	225
2	1.0	238
3	1.5	240
4	2.0	248
5	2.5	260
6	3.0	265
7	3.5	283
8	4.0	301
9	4.5	305
10	5.0	309