机器学习树叶分类与聚类

1导入包

import os
import matplotlib.image as img
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import warnings# 备份原有warnings过滤器
filters = warnings.filters[:]
# 新增一条忽略DeprecationWarning的过滤规则
warnings.simplefilter('ignore', DeprecationWarning)
# import sets
# # 恢复原来的过滤器
# warnings.filters = filters

2查看数据

save_path='./filtered_imgs'
if os.path.exists(save_path) is False:os.makedirs(save_path)

os.listdir()

['.ipynb_checkpoints','data_imgs','data_imgs2','filtered_imgs','images','ML_lesson10_KMeans聚类.ipynb','notebook.tex','render.html','sample_submission.csv','test.csv','train.csv','Untitled.ipynb','Untitled1.ipynb','树叶分类.html','树叶分类实现.ipynb','聚类.ipynb','聚类测试.ipynb']

# 对文件名称重新排序
img_path='./images'
img_name_list=os.listdir(img_path)
img_name_list.sort(key=lambda x: int(x.split('.')[0]))

##顺序读取前12张图片并排序
DImage=[]
for img_name in img_name_list[:12]:img_full_path=os.path.join(img_path, img_name)DImage.append(img.imread(img_full_path))

plt.style.use('ggplot')

##可视化树叶图片
f=plt.figure(figsize=(8,6))
for i in range(12):plt.subplot(3,4,i+1)plt.axis("off")plt.title("image_ID:{0}".format(img_name_list[i].split('.jpg')[0]))plt.imshow(DImage[i],cmap='hot')
plt.show()

3读取训练集和测试集

Train = pd.read_csv('train.csv')
Train_id = Train['id']
Test = pd.read_csv('test.csv')
Test_id = Test['id']
Test.drop(['id'],inplace = True, axis = 1)

# 查看训练集数据描述
Train.describe()

	id	margin1	margin2	margin3	margin4	margin5	margin6	margin7	margin8	margin9	...	texture55	texture56	texture57	texture58	texture59	texture60	texture61	texture62	texture63	texture64
count	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	...	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000	990.000000
mean	799.595960	0.017412	0.028539	0.031988	0.023280	0.014264	0.038579	0.019202	0.001083	0.007167	...	0.036501	0.005024	0.015944	0.011586	0.016108	0.014017	0.002688	0.020291	0.008989	0.019420
std	452.477568	0.019739	0.038855	0.025847	0.028411	0.018390	0.052030	0.017511	0.002743	0.008933	...	0.063403	0.019321	0.023214	0.025040	0.015335	0.060151	0.011415	0.039040	0.013791	0.022768
min	1.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
25%	415.250000	0.001953	0.001953	0.013672	0.005859	0.001953	0.000000	0.005859	0.000000	0.001953	...	0.000000	0.000000	0.000977	0.000000	0.004883	0.000000	0.000000	0.000000	0.000000	0.000977
50%	802.500000	0.009766	0.011719	0.025391	0.013672	0.007812	0.015625	0.015625	0.000000	0.005859	...	0.004883	0.000000	0.005859	0.000977	0.012695	0.000000	0.000000	0.003906	0.002930	0.011719
75%	1195.500000	0.025391	0.041016	0.044922	0.029297	0.017578	0.056153	0.029297	0.000000	0.007812	...	0.043701	0.000000	0.022217	0.009766	0.021484	0.000000	0.000000	0.023438	0.012695	0.029297
max	1584.000000	0.087891	0.205080	0.156250	0.169920	0.111330	0.310550	0.091797	0.031250	0.076172	...	0.429690	0.202150	0.172850	0.200200	0.106450	0.578130	0.151370	0.375980	0.086914	0.141600

8 rows × 193 columns

Train.head()

	id	species	margin1	margin2	margin3	margin4	margin5	margin6	margin7	margin8	...	texture55	texture56	texture57	texture58	texture59	texture60	texture61	texture62	texture63	texture64
0	1	Acer_Opalus	0.007812	0.023438	0.023438	0.003906	0.011719	0.009766	0.027344	0.0	...	0.007812	0.000000	0.002930	0.002930	0.035156	0.0	0.0	0.004883	0.000000	0.025391
1	2	Pterocarya_Stenoptera	0.005859	0.000000	0.031250	0.015625	0.025391	0.001953	0.019531	0.0	...	0.000977	0.000000	0.000000	0.000977	0.023438	0.0	0.0	0.000977	0.039062	0.022461
2	3	Quercus_Hartwissiana	0.005859	0.009766	0.019531	0.007812	0.003906	0.005859	0.068359	0.0	...	0.154300	0.000000	0.005859	0.000977	0.007812	0.0	0.0	0.000000	0.020508	0.002930
3	5	Tilia_Tomentosa	0.000000	0.003906	0.023438	0.005859	0.021484	0.019531	0.023438	0.0	...	0.000000	0.000977	0.000000	0.000000	0.020508	0.0	0.0	0.017578	0.000000	0.047852
4	6	Quercus_Variabilis	0.005859	0.003906	0.048828	0.009766	0.013672	0.015625	0.005859	0.0	...	0.096680	0.000000	0.021484	0.000000	0.000000	0.0	0.0	0.000000	0.000000	0.031250

5 rows × 194 columns

Test.head()

	margin1	margin2	margin3	margin4	margin5	margin6	margin7	margin8	margin9	margin10	...	texture55	texture56	texture57	texture58	texture59	texture60	texture61	texture62	texture63	texture64
0	0.019531	0.009766	0.078125	0.011719	0.003906	0.015625	0.005859	0.0	0.005859	0.023438	...	0.006836	0.000000	0.015625	0.000977	0.015625	0.0	0.0	0.000000	0.003906	0.053711
1	0.007812	0.005859	0.064453	0.009766	0.003906	0.013672	0.007812	0.0	0.033203	0.023438	...	0.000000	0.000000	0.006836	0.001953	0.013672	0.0	0.0	0.000977	0.037109	0.044922
2	0.000000	0.000000	0.001953	0.021484	0.041016	0.000000	0.023438	0.0	0.011719	0.005859	...	0.128910	0.000000	0.000977	0.000000	0.000000	0.0	0.0	0.015625	0.000000	0.000000
3	0.000000	0.000000	0.009766	0.011719	0.017578	0.000000	0.003906	0.0	0.003906	0.001953	...	0.012695	0.015625	0.002930	0.036133	0.013672	0.0	0.0	0.089844	0.000000	0.008789
4	0.001953	0.000000	0.015625	0.009766	0.039062	0.000000	0.009766	0.0	0.005859	0.000000	...	0.000000	0.042969	0.016602	0.010742	0.041016	0.0	0.0	0.007812	0.009766	0.007812

5 rows × 192 columns

Train['species'].value_counts().head()

Quercus_Agrifolia               10
Quercus_Chrysolepis             10
Alnus_Cordata                   10
Viburnum_x_Rhytidophylloides    10
Ginkgo_Biloba                   10
Name: species, dtype: int64

print("树叶种类数目为：",len(set(Train['species'])))

树叶种类数目为： 99

## 把species转换为类标签。如把树叶名转化为数字标签
map_dic = {}
i =- 1
for _ in Train['species']:if _ in map_dic:passelse:i+=1map_dic[_] = map_dic.get(_, i)

[(key, value) for key, value in map_dic.items()][:5]

[('Acer_Opalus', 0),('Pterocarya_Stenoptera', 1),('Quercus_Hartwissiana', 2),('Tilia_Tomentosa', 3),('Quercus_Variabilis', 4)]

len(map_dic)

99

Train['species'].replace(map_dic.keys(), map_dic.values(), inplace=True)

Train.drop(['id'], inplace = True, axis = 1)

Train_ture = Train['species']

3.1画出相关性矩阵（需要根据相关性矩阵，选择特征进行特征工程）

corr = Train.corr()
f, ax = plt.subplots(figsize=(25, 25))
cmap = sns.diverging_palette(220, 10, as_cmap=True)
sns.heatmap(corr, cmap=cmap, vmax=.3, center=0,square=True, linewidths=.5)
plt.show()

# 判断是否存在缺失值，若为False，则无缺失值
np.all(np.any(pd.isnull(Train)))

False

X = Train.drop(['species'], axis=1)
y = Train['species']
print(y.head())
print("训练集尺寸:", X.shape)

0    0
1    1
2    2
3    3
4    4
Name: species, dtype: int64
训练集尺寸: (990, 192)

## 划分训练集和测试集
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X.values, y.values, test_size=0.2, random_state=42)
print("训练集数据尺寸",X_train.shape)
print("测试集数据尺寸",X_test.shape)
print("训练集目标尺寸",y_train.shape)
print("测试集目标尺寸",y_test.shape)

训练集数据尺寸 (792, 192)
测试集数据尺寸 (198, 192)
训练集目标尺寸 (792,)
测试集目标尺寸 (198,)

3.2数据标准化

from sklearn.preprocessing import StandardScaler

# 数据标准化
standerScaler = StandardScaler()
X_train = standerScaler.fit_transform(X_train)
X_test = standerScaler.transform(X_test)

4讨论是否需要PCA

4.1没有经过PCA降维的KNN

%%time
from sklearn.preprocessing import StandardScalerX_train_shape1 = X_train.shape
X_test_shape1 = X_test.shape
print(X_train.shape[1])

192
Wall time: 0 ns

%%time
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_scoreclf=KNeighborsClassifier(2)
clf.fit(X_train, y_train)
train_predictions = clf.predict(X_test)
notpca_score = accuracy_score(y_test, train_predictions)
print(notpca_score)

0.9696969696969697
Wall time: 134 ms

4.2经过PCA降维过的KNN

from sklearn.decomposition import PCApca = PCA(n_components=0.95)
X_train = pca.fit_transform(X_train)
X_test = pca.transform(X_test)
X_train_shape2 = X_train.shape
X_test_shape2 = X_test.shape
print(X_train.shape[1])

68

%%timeclf=KNeighborsClassifier(2)
clf.fit(X_train, y_train)
train_predictions = clf.predict(X_test)
pca_score = accuracy_score(y_test, train_predictions)
print(pca_score)

0.9646464646464646
Wall time: 34.9 ms

4.3比较降维前后

data_score = pd.DataFrame([[X_train_shape1, X_train_shape2],[X_test_shape1, X_test_shape2], [notpca_score, pca_score]])
# 添加中文行索引
data_score.index = ["X_train形状", "X_test形状", "Accuracy得分"]
# 添加中文列索引
data_score.columns = ["降维前", "降维后"]
print("KNN")
data_score

	降维前	降维后
X_train形状	(792, 192)	(792, 68)
X_test形状	(198, 192)	(198, 68)
Accuracy得分	0.969697	0.964646

讨论: pca降维在保留95%的信息下，与降维前的准确率相比略低，相差不大的情况下，以获得较快的计算速度。

5讨论分类算法优劣

score_classify_list = []
model_classify_list = []
model_predict_list = []

5.1KNN

from sklearn.neighbors import KNeighborsClassifierknn_clf0 = KNeighborsClassifier()
knn_clf0.fit(X_train, y_train)
print("*"*30)
print('KNeighborsClassifier')y_predict = knn_clf0.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("KNN")
model_predict_list.append(y_predict)

运行结果:
******************************
KNeighborsClassifier
Accuracy: 97.9798%

5.2KNN网格搜索优化

from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import GridSearchCVparam_grid = [{'weights':["uniform"],'n_neighbors':[i for i in range(2, 4)]},{'weights':["distance"],'n_neighbors':[i for i in range(2, 4)],'p':[i for i in range(1, 3)]}
]knn_clf = KNeighborsClassifier()
gird_search = GridSearchCV(knn_clf, param_grid)
gird_search.fit(X_train, y_train)
print("最佳超参数: ", gird_search.best_params_)
knn_clf = gird_search.best_estimator_
print("*"*30)
print('KNeighborsClassifier网格搜索')y_predict = knn_clf.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("KNN网格搜索")
model_predict_list.append(y_predict)

运行结果:
最佳超参数: {‘n_neighbors’: 2, ‘p’: 1, ‘weights’: ‘distance’}
******************************
KNeighborsClassifier网格搜索
Accuracy: 98.9899%

5.3SVC

from sklearn.svm import SVC
svc_clf = SVC(probability=True)
svc_clf.fit(X_train, y_train)print("*"*30)
print('SVC')y_predict = svc_clf.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("SVC")
model_predict_list.append(y_predict)

运行结果:
******************************
SVC
Accuracy: 97.4747%

5.4逻辑回归

from sklearn.linear_model import LogisticRegressionCVlr = LogisticRegressionCV(multi_class="ovr", fit_intercept=True, Cs=np.logspace(-2,2,20), cv=2, penalty="l2", solver="lbfgs", tol=0.01)lr.fit(X_train,y_train)print("*"*30)
print('逻辑回归')y_predict = lr.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("逻辑回归")
model_predict_list.append(y_predict)

运行结果:
******************************
逻辑回归
Accuracy: 98.9899%

5.5voting

warnings.simplefilter('ignore', DeprecationWarning)from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.ensemble import VotingClassifiervoting_clf = VotingClassifier(estimators=[('log_clf', LogisticRegression()),('svm_clf', SVC(probability=True)),
], voting='hard')voting_clf.fit(X_train,y_train)print("*"*30)
print('voting')y_predict = voting_clf.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("voting")
model_predict_list.append(y_predict)

运行结果:
******************************
voting
Accuracy: 97.4747%

5.6随机森林

from sklearn.ensemble import RandomForestClassifierrf_clf = RandomForestClassifier(n_estimators=250,random_state=666, oob_score=True)
rf_clf.fit(X_train, y_train)print("*"*30)
print('随机森林')y_predict = rf_clf.predict(X_test)
score = accuracy_score(y_test, y_predict)
print("Accuracy: {:.4%}".format(score))score_classify_list.append(score)
model_classify_list.append("随机森林")
model_predict_list.append(y_predict)

运行结果:
******************************
随机森林
Accuracy: 96.9697%

5.7比较分类算法

data_score = pd.DataFrame(score_classify_list)
models = model_classify_list
features = ["accuracy_score得分"]
# 添加中文行索引
data_score.index = models
# 添加中文列索引
data_score.columns = features
data_score

不同分类算法的得分:

	accuracy_score得分
KNN	0.979798
KNN网格搜索	0.989899
SVC	0.974747
逻辑回归	0.989899
voting	0.974747
随机森林	0.969697

from pyecharts import options as opts
from pyecharts.charts import Bar
from pyecharts.globals import ThemeTypemodels_score = [round(i, 4) for i in score_classify_list]def bar_reversal_axis() -> Bar:c = (Bar(init_opts=opts.InitOpts(width="800px",height="300px", theme=ThemeType.WONDERLAND)).add_xaxis(models).add_yaxis("model", models_score).reversal_axis().set_series_opts(label_opts=opts.LabelOpts(position="right")).set_global_opts(title_opts=opts.TitleOpts(title="分类算法", subtitle="得分")))return c
c = bar_reversal_axis()
c.render("分类算法得分.html")
c.render_notebook()

可视化不同分类算法的得分:

# (6, 198)
model_predict_array = np.array(model_predict_list)
test_id_list = ["test_"+str(i) for i in range(model_predict_array.shape[1])]def bar_datazoom_slider() -> Bar:c = (Bar(init_opts=opts.InitOpts(width="1000px",height="200px",theme=ThemeType.WONDERLAND)).add_xaxis(test_id_list[:]).add_yaxis(models[0], model_predict_array[0,:].tolist()).add_yaxis(models[1], model_predict_array[1,:].tolist()).add_yaxis(models[2], model_predict_array[2,:].tolist()).add_yaxis(models[3], model_predict_array[3,:].tolist()).add_yaxis(models[4], model_predict_array[4,:].tolist()).add_yaxis(models[5], model_predict_array[5,:].tolist()).set_global_opts(title_opts=opts.TitleOpts(title="分类", subtitle="预测类别"),datazoom_opts=[opts.DataZoomOpts(type_="slider")]))return c

c = bar_datazoom_slider()
c.render("分类预测类别.html")
c.render_notebook()

可视化不同分类算法的预测结果:

warnings.simplefilter('ignore', DeprecationWarning)
def plot_learning_curve(model, x_start, x_shop, metrics, X_train_xx, X_test_xx, y_train_xx, y_test_xx,model_name):train_score = []test_score = []for i in range(x_start, x_shop, 100):model.fit(X_train_xx[:i], y_train_xx[:i])y_train_predict = model.predict(X_train_xx[:i])train_score.append(metrics(y_train_xx[:i], y_train_predict))y_test_predict = model.predict(X_test_xx)test_score.append(metrics(y_test_xx, y_test_predict))plt.plot([i for i in range(x_start, x_shop, 100)], np.sqrt(train_score), label="train")plt.plot([i for i in range(x_start, x_shop, 100)], np.sqrt(test_score), label="test")plt.legend()plt.title(model_name)plt.xlabel('训练数量')plt.ylabel('正确率')plt.axis([x_start, x_shop, 0., 1.1])

X_train_xx = X_train.copy()
X_test_xx = X_test.copy()
y_train_xx = y_train.copy()
y_test_xx = y_test.copy()

%%time
model_clf_list = [knn_clf0, knn_clf, svc_clf, lr, voting_clf, rf_clf]
figure, axes = plt.subplots(nrows=4, ncols=2, figsize=(15, 25), dpi=80)
warnings.simplefilter('ignore', Warning)
for i in range(1, 7):plt.subplot(3,2,i)plot_learning_curve(model_clf_list[i-1], 22, len(X_train_xx)+1, accuracy_score, X_train_xx, X_test_xx, y_train_xx, y_test_xx, models[i-1])
plt.show()

学习曲线:观察学习曲线，这6种模型的最终准确率都非常接近于100%，KNN网格搜索和voting集成学习的训练结果和测试结果比较接近，可知两者的过拟合程度非常小。又因为KNN网格搜索最高，所以KNN网格搜索训练出的模型最为理想。

Wall time: 2min 21s

6预测树叶图片进行分类

6.1测试集，预测结果99种类别

### warnings.simplefilter('ignore', DeprecationWarning)
# 测试集，预测结果99种类别
Test_sta = standerScaler.transform(Test)
Test_pca = pca.transform(Test_sta)
Test_predict = knn_clf.predict(Test_pca)
print("预测结果尺寸为:")
Test_predict.shape

预测结果尺寸为:
(594,)

6.2初始化Test_label_dic， Train_label_dic

Test_label_dic={}
for i in range(99):Test_label_dic[i]=np.where(Test_predict==i)[0]Train_label_dic={}
for i in range(99):Train_label_dic[i]=np.where(Train_ture==i)[0]

6.3建立所有图片的数据集

DImage=[]
for img_name in img_name_list:img_full_path=os.path.join(img_path, img_name)DImage.append(img.imread(img_full_path))
train_y = {}
test_y = {}

6.4生成分类图片

# import pysnooper# @pysnooper.snoop(r'C:\Users\linyihua\Desktop\mylog/file.log')
def imagesclassifier(DImage,Train_id,Test_id,Train_label_dic,Test_label_dic,root_path, train_y={}, test_y={}):'''DImage为图像数据集(循环读取的1584张图片)，Train_id,Test_id分别是train.csv和test.csv 内id列，Train_label_dic,Test_label_dic是树叶类别标签和图片索引对应关系'''if os.path.exists(root_path) is False:os.makedirs(root_path)save_path=root_pathfor i in range(99):train_val=Train_id.values[np.array(Train_label_dic[i]).reshape(-1)]-1test_val=Test_id.values[np.array(Test_label_dic[i]).reshape(-1)]-1train_y[i] = train_valtest_y[i] = test_valTrain_imgs=np.array(DImage)[train_val]Test_imgs=np.array(DImage)[test_val]for index, _ in enumerate(Train_imgs):img_name='train'+str(train_val[index])+'.jpg'save_path=os.path.join(save_path,str(i))if os.path.exists(save_path) is False:os.makedirs(save_path)img.imsave(os.path.join(save_path,img_name),_,cmap='binary')save_path=root_pathfor index, _ in enumerate(Test_imgs):img_name='test'+str(test_val[index])+'.jpg'save_path=os.path.join(save_path,str(i))if os.path.exists(save_path) is False:os.makedirs(save_path)img.imsave(os.path.join(save_path,img_name),_,cmap='binary')save_path=root_pathreturn train_y, test_yprint(train_val)

%%time
train_y, test_y = imagesclassifier(DImage,Train_id,Test_id,Train_label_dic,Test_label_dic,save_path)

Wall time: 1min 16s

6.5查看在filtered_imgs下生成99个文件夹(文件夹4内的图片如下)

##顺序读取前12张图片并排序
img_name_list1=os.listdir('./filtered_imgs/0')
img_path1='./filtered_imgs/0'
DImage1=[]
for img_name1 in img_name_list1[:12]:img_full_path1=os.path.join(img_path1,img_name1)DImage1.append(img.imread(img_full_path1))
##可视化树叶图片
f=plt.figure(figsize=(15,6))
for i in range(12):plt.subplot(3,4,i+1)plt.axis("off")plt.title("image_ID:{0}".format(img_name_list1[i].split('.jpg')[0]))plt.imshow(DImage1[i],cmap='hot')
plt.show()

6.6查看在filtered_imgs下生成99个文件夹(文件夹7内的图片如下)

##顺序读取前12张图片并排序
img_name_list2=os.listdir('./filtered_imgs/7')
img_path2='./filtered_imgs/7'
DImage2=[]
for img_name2 in img_name_list2[:12]:img_full_path2=os.path.join(img_path2,img_name2)DImage2.append(img.imread(img_full_path2))
##可视化树叶图片
f=plt.figure(figsize=(15, 10))
for i in range(12):plt.subplot(3,4,i+1)plt.axis("off")plt.title("image_ID:{0}".format(img_name_list2[i].split('.jpg')[0]))plt.imshow(DImage2[i],cmap='hot')
plt.show()

7不同聚类分析

7.1导入数据

# 导入数据
Train2 = pd.read_csv('train.csv')
Test2 = pd.read_csv('test.csv')
print(Train2.shape, Test2.shape)

(990, 194) (594, 193)

7.2数据处理

## 数据处理
Train2.drop(['species'],inplace = True, axis = 1)
data = np.concatenate((Train2,Test2), axis=0)
data = pd.DataFrame(data)
columns = Test2.columns
data.columns = columns
data = data.sort_values(by="id", ascending=True)
data.head()

	id	margin1	margin2	margin3	margin4	margin5	margin6	margin7	margin8	margin9	...	texture55	texture56	texture57	texture58	texture59	texture60	texture61	texture62	texture63	texture64
0	1.0	0.007812	0.023438	0.023438	0.003906	0.011719	0.009766	0.027344	0.0	0.001953	...	0.007812	0.000000	0.002930	0.002930	0.035156	0.0	0.0	0.004883	0.000000	0.025391
1	2.0	0.005859	0.000000	0.031250	0.015625	0.025391	0.001953	0.019531	0.0	0.000000	...	0.000977	0.000000	0.000000	0.000977	0.023438	0.0	0.0	0.000977	0.039062	0.022461
2	3.0	0.005859	0.009766	0.019531	0.007812	0.003906	0.005859	0.068359	0.0	0.000000	...	0.154300	0.000000	0.005859	0.000977	0.007812	0.0	0.0	0.000000	0.020508	0.002930
990	4.0	0.019531	0.009766	0.078125	0.011719	0.003906	0.015625	0.005859	0.0	0.005859	...	0.006836	0.000000	0.015625	0.000977	0.015625	0.0	0.0	0.000000	0.003906	0.053711
3	5.0	0.000000	0.003906	0.023438	0.005859	0.021484	0.019531	0.023438	0.0	0.013672	...	0.000000	0.000977	0.000000	0.000000	0.020508	0.0	0.0	0.017578	0.000000	0.047852

5 rows × 193 columns

X = data.iloc[:, 1:]
y_id = data["id"]

X.shape

(1584, 192)

7.2.1数据标准化

from sklearn.preprocessing import StandardScalerstanderScaler2 = StandardScaler()
X = standerScaler2.fit_transform(X)

7.2.2pca，降为2维方便可视化

from sklearn.decomposition import PCApca2 = PCA(n_components=2)
X_reduction = pca2.fit_transform(X)

y_predict_list = []

7.3scatter_cluster(cluster_num, plt, model_name, X_reduction, y_predict):

def scatter_cluster(cluster_num, plt, model_name, X_reduction, y_predict):colors = ["#64A600","#A6A600", "#C6A300","#EA7500", "#AD5A5A","#A5A552","#5CADAD","#8080C0","#EA0000","#FF359A","#D200D2","#9F35FF","#2828FF","#0080FF","#00CACA","#02DF82",]markers = ["o", "^", "s", "p", "x", "+", "d", "*"] * 2#     plt.figure(figsize=(), dpi=80)plt.grid(linestyle="--", alpha=0.5)for i in range(cluster_num):plt.scatter(X_reduction[y_predict==i, 0], X_reduction[y_predict==i, 1],color=colors[i], marker=markers[i], label=str(i))plt.title(model_name+":聚类数量为"+str(cluster_num))plt.legend()
#     plt.show()

model_names = []

7.4KMeans

from sklearn.cluster import KMeans# km = KMeans()
# km.fit(X_reduction, y)
# y_predict = km.predict(X_reduction)
from sklearn.cluster import KMeans
km = KMeans(n_clusters=8,init='k-means++',n_init=10,max_iter=300,tol=0.0001,precompute_distances='auto',verbose=0,random_state=None,copy_x=True,n_jobs=1,algorithm='auto')
#n_clusters:class的个数；
#max_inter:每一个初始化值下，最大的iteration次数；
#n_init:尝试用n_init个初始化值进行拟合;
#tol:within-cluster sum of square to declare convergence;
#init=‘k-means++’：可使初始化的centroids相互远离；
km.fit(X_reduction, y)
y_predict = km.predict(X_reduction)
y_predict_list.append(y_predict)
model_names.append("KMeans")

7.5Birch

warnings.simplefilter('ignore', FutureWarning)from sklearn.cluster import Birchy_predict = Birch(n_clusters = 8).fit_predict(X_reduction)
y_predict_list.append(y_predict)
model_names.append("Birch")

7.6MiniBatchKMeans

from sklearn.cluster import MiniBatchKMeansy_predict = MiniBatchKMeans(n_clusters = 8).fit_predict(X_reduction)
y_predict_list.append(y_predict)
model_names.append("MiniBatchKMeans")

7.7高斯混合聚类

from sklearn import mixturey_predict = mixture.GMM(n_components=8).fit_predict(X_reduction)
y_predict_list.append(y_predict)
model_names.append("高斯混合聚类")

7.8比较不同聚类的效果

figure, axes = plt.subplots(nrows=2, ncols=2, figsize=(15, 12), dpi=80)
cluster_num = 8
for i in range(1, 5):plt.subplot(2,2,i)scatter_cluster(cluster_num, plt, model_names[i-1], X_reduction, y_predict_list[i-1])
# 添加网格显示
plt.grid(linestyle="--", alpha=0.8)
plt.show()

不同聚类可视化:将特征运用PCA降维，降成两维以方便可视化，根据簇内距离最近，簇间距离最大原则，KMeans比较理想。

8分析聚类聚多少类合适

def KMeans_clusters(clusters_num, X):km = KMeans(n_clusters=clusters_num,init='k-means++',n_init=10,max_iter=300,tol=0.0001,precompute_distances='auto',verbose=0,random_state=None,copy_x=True,n_jobs=1,algorithm='auto')km.fit(X, y)y_predict = km.predict(X)return y_predictcluster_num = 4
cluster_name = "KMeans"
figure, axes = plt.subplots(nrows=2, ncols=2, figsize=(15, 12), dpi=80)
for i in range(1, 5):y_predict = KMeans_clusters(cluster_num*i, X_reduction)plt.subplot(2,2,i)scatter_cluster(cluster_num*i, plt, cluster_name, X_reduction, y_predict)# 添加网格显示
plt.grid(linestyle="--", alpha=0.8)
plt.show()

可视化不同簇数聚类:观察可知分成16簇比较理性。

9聚类结果

9.1imagesclassifier2(cluster_num, DImage2, y_predict_KMeans2 ,root_path):

# 聚16类
cluster_num = 16
y_predict = KMeans_clusters(cluster_num, X)def imagesclassifier2(cluster_num, DImage, y_predict ,root_path):if os.path.exists(root_path) is False:os.makedirs(root_path)save_path=root_pathfor i in range(cluster_num):data_val=np.where(y_predict==i)[0]data_imgs=np.array(DImage)[data_val]for index, _ in enumerate(data_imgs):img_name='data'+str(data_val[index])+'.jpg'save_path=os.path.join(save_path,str(i))if os.path.exists(save_path) is False:os.makedirs(save_path)img.imsave(os.path.join(save_path,img_name),_,cmap='binary')save_path=root_path

imagesclassifier2(cluster_num, DImage, y_predict, r"./data_imgs")

9.2顺序读取前12张图片并排序

##顺序读取前12张图片并排序
img_name_list22=os.listdir('./data_imgs/5')
img_path22='./data_imgs/5'
DImage22=[]
for img_name22 in img_name_list22[:12]:img_full_path22=os.path.join(img_path22,img_name22)DImage22.append(img.imread(img_full_path22))
##可视化树叶图片
f=plt.figure(figsize=(15, 10))
for i in range(12):plt.subplot(3,4,i+1)plt.axis("off")plt.title("image_ID:{0}".format(img_name_list22[i].split('.jpg')[0]))plt.imshow(DImage22[i],cmap='hot')
plt.show()

机器学习树叶分类与聚类相关推荐

1. 机器学习中分类与聚类的本质区别

   机器学习中分类与聚类的本质区别 机器学习中有两类的大问题,一个是分类,一个是聚类. 在我们的生活中,我们常常没有过多的去区分这两个概念,觉得聚类就是分类,分类也差不多就是聚类,下面,我们就具体来研究下 ...

2. 2021-02-02美赛前MATLAB的学习笔记(机器学习（分类、聚类、深度学习）)

   机器学习 机器学习是一中工具.方法,通过对机器训练,进而学习到某种规律或者模式,并建立预测未来结果的模型. 机器学习可以分为监督学习和无监督学习 有监督学习方法,是提供答案的,主要包括分类和回归 无监 ...

3. 回归、分类与聚类：三大方向剖解机器学习算法的优缺点

   回归.分类与聚类:三大方向剖解机器学习算法的优缺点 2017-05-20 13:56:14    机器学习   数学    3 0 0 在本教程中,作者对现代机器学习算法进行一次简要的实战梳理.虽然类 ...

4. Py之scikit-learn：机器学习sklearn库的简介、六大基本功能介绍(数据预处理/数据降维/模型选择/分类/回归/聚类)、安装、使用方法(实际问题中如何选择最合适的机器学习算法)之详细攻略

   Py之scikit-learn:机器学习sklearn库的简介(组件/版本迭代).六大基本功能介绍(数据预处理/数据降维/模型选择/分类/回归/聚类).安装.使用方法(实际问题中如何选择最合适的机器学 ...

5. AI：人工智能领域算法思维导图集合之有监督学习/无监督学习/强化学习类型的具体算法简介(预测函数/优化目标/求解算法)、分类/回归/聚类/降维算法模型选择思路、11类机器学习算法详细分类之详细攻略

   AI:人工智能领域算法思维导图集合之有监督学习/无监督学习/强化学习类型的具体算法简介(预测函数/优化目标/求解算法).分类/回归/聚类/降维算法模型选择思路.11类机器学习算法详细分类(决策树/贝叶 ...

6. 机器学习四大基本模型：回归、分类、聚类、降维

   在本文中,我对现代机器学习算法进行了简要梳理,我通过查阅转载众多博客和资料,基于实践中的经验,讨论每个算法的优缺点,并以机器学习入门者的角色来看待各个模型. 主要内容来自<机器之心>:回归 ...

7. 二分类最优阈值确定_机器学习-分类和聚类

   ## 机器学习-分类和聚类.分类和回归.逻辑回归和KNN 分类和聚类的概念: ** 1.分类:使用已知的数据集(训练集)得到相应的模型,通过这个模型可以划分未知数据.分类涉及到的数据集通常是带有标签 ...

8. 收藏！机器学习算法分类图谱及其优缺点综合分析

   来源:必达智库 近日,Coggle对各类机器学习算法进行了归纳整理,形成了一个较为完整的机器学习算法分类图谱,并对每一类算法的优缺点进行了分析.具体分类如下: 正则化算法(Regularization ...

9. [Python从零到壹] 十四.机器学习之分类算法五万字总结全网首发（决策树、KNN、SVM、分类对比实验）

   欢迎大家来到"Python从零到壹",在这里我将分享约200篇Python系列文章,带大家一起去学习和玩耍,看看Python这个有趣的世界.所有文章都将结合案例.代码和作者的经验讲 ...

最新文章

1. makefille的使用
2. RHEL6.5/Centos6.5 搭建bugzilla
3. 虚拟语气和推测（三）
4. PAT乙级 1086 就不告诉你 （附测试点1，2排查及用例）
5. python数据库操作常用功能使用详解(创建表/插入数据/获取数据)
6. python 写文件 编码_Python文件写入时的编码问题解决
7. ABViewer免费汉化下载注册地址图形查看器教程功能介绍
8. 2015年度夏季假期学习内容
9. 阿里云携手蓝凌软件，打造全球化企业智慧办公平台
10. java架构师一般多少岁,大量教程
11. Win10 KeilC51-C251-ARM共存方法
12. S3C2440裸机------GPIO
13. 核函数是什么-数据的升维与降维
14. MySql Order By 多个字段 排序规则
15. Java什么是对象？
16. 【TJOI 2015】弦论
17. 在win7系统中用U盘安装Ubuntu16.04
18. EBS 销售订单行单条一次或多次发运确认API（wsh_new_delivery_actions.confirm_delivery）详解
19. 3D游戏编程与设计作业6-Unity实现打飞碟游戏改进版（Hit UFO）
20. 蓝牙连接每次弹出确认框问题的排查及解决

热门文章

1. 史上最直白易懂的五险一金介绍，初入职场的新人必知必懂！
2. 6亿用户音乐推荐系统，我们这样做！
3. 论据和论证过程比论点更重要
4. MODBUS RTU通讯常见错误代码
5. 学生的认知风格类型有哪些_认知风格的四种类型
6. FTP无法连接解决办法
7. 远程计算机guest密码更改了,怎么给guest设置密码
8. 《仙剑奇侠传3》流程攻略2
9. 站在原地就是退步——除了死磕通道，云通讯服务商还该做些什么？
10. 线性回归，波斯顿房价的预测