Digit Recognizer 练习地址

相关博文：[Hands On ML] 3. 分类（MNIST手写数字预测）

1. Baseline KNN

• 读取数据

import pandas as pd
train = pd.read_csv('train.csv')
X_test = pd.read_csv('test.csv')

• 特征、标签分离

train.head()
y_train = train['label']
X_train = train.drop(['label'], axis=1)
X_train

• 网格搜索 KNN 模型最佳参数

from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score
# help(KNeighborsClassifier)
para_dict = [{'weights':["uniform", "distance"], 'n_neighbors':[3,4,5], 'leaf_size':[10,20]}
]
knn_clf = KNeighborsClassifier()
grid_search = GridSearchCV(knn_clf, para_dict, cv=3,scoring='accuracy',n_jobs=-1)
grid_search.fit(X_train, y_train)

输出
GridSearchCV(cv=3, estimator=KNeighborsClassifier(), n_jobs=-1,param_grid=[{'leaf_size': [10, 20], 'n_neighbors': [3, 4, 5],'weights': ['uniform', 'distance']}],scoring='accuracy')

• 最佳参数

grid_search.best_params_
# {'leaf_size': 10, 'n_neighbors': 4, 'weights': 'distance'}

• 最好得分

grid_search.best_score_
# 0.9677619047619048

• 生成 test 集预测结果

y_pred = grid_search.predict(X_test)

• 写入结果文件

image_id = pd.Series(range(1,len(y_pred)+1))
output = pd.DataFrame({'ImageId':image_id, 'Label':y_pred})
output.to_csv("submission.csv",  index=False) # 不要index列

• 预测结果

排行榜

以上 KNN 模型得分 0.97067，目前排名2467

2. Try SVC

• 读取数据

import pandas as pd
train = pd.read_csv('train.csv')
X_test = pd.read_csv('test.csv')
y_train = train['label']
X_train = train.drop(['label'], axis=1)

• 导入包

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.svm import SVC, LinearSVC
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score

• 搜索最佳参数

pipeline = Pipeline([("scaler",StandardScaler()),('clf', SVC(decision_function_shape="ovr", gamma="auto"))
])from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import reciprocal, uniformparam_distributions = {"clf__gamma": reciprocal(0.001, 0.1), "clf__C": uniform(1, 10)}
rnd_search_cv = RandomizedSearchCV(pipeline, param_distributions, n_iter=10, verbose=2, cv=3)rnd_search_cv.fit(X_train, y_train)

• 训练花费12个小时 [Parallel(n_jobs=1)]: Done 30 out of 30 | elapsed: 744.1min finished

rnd_search_cv.best_estimator_

• 最佳评估器

Pipeline(steps=[('scaler', StandardScaler()),('clf',SVC(C=10.729327185542381, gamma=0.0022750096640207287))])

• 最好得分

rnd_search_cv.best_score_
# 0.9584285714285713

• 预测

y_pred = rnd_search_cv.best_estimator_.predict(X_test)
image_id = pd.Series(range(1,len(y_pred)+1))
output = pd.DataFrame({'ImageId':image_id, 'Label':y_pred})
output.to_csv("submission_svc.csv",  index=False)

SVC 支持向量机分类模型 得分 0.96464 没有上面 KNN 模型高（KNN 得分 0.97067）