from Data import DataExtract
X_train, X_test, Y_train, Y_test = DataExtract.load_minist_csv()
# 测试 LogisticRegression RandomForest xgboost
from sklearn.linear_model import LogisticRegressionCV
from sklearn.ensemble import RandomForestClassifier
from xgboost import XGBClassifier
models = [
('LogisticRegression', LogisticRegressionCV(Cs=10,
cv=5)), # Cs正则约束，越小越强 cv:交叉验证
('RandomForest',
RandomForestClassifier(n_estimators=50)), # , criterion='gini'
('XGBoost',
XGBClassifier(max_depth=3,
n_estimators=100,
silent=True,
objective='multi:softmax'))
]
for name, model in models:
model.fit(X_train, Y_train)
print(name, '训练集正确率：', model.score(X_train, Y_train))
print(name, '测试集正确率：', model.score(X_test, Y_test))
# LogisticRegression 训练集正确率： 0.9294345238095238
# LogisticRegression 测试集正确率： 0.9158333333333334
# RandomForest 训练集正确率： 1.0
# RandomForest 测试集正确率： 0.8848809523809524
# XGBoost 训练集正确率： 0.9670238095238095
# XGBoost 测试集正确率： 0.9277380952380953
import numpy as np
from datetime import datetime
from Data import DataExtract, DataTransform
from Minist.Common.Util import error_rate
from sklearn.utils import shuffle
# keras test error rate 0.036071.
# Elapsted time for keras rmsprop: 0:00:25.107859
# tensorflow test error rate 0.025238.
# Elapsted time for tensorflow rmsprop: 0:08:09.270659
X_train, X_test, Y_train, Y_test = DataExtract.load_minist_csv(
pca=False) # pca=False
class_num = 10
Y_train_onehot = DataTransform.y2one_hot(Y_train, class_num=class_num)
Y_test_onehot = DataTransform.y2one_hot(Y_test, class_num=class_num)
N, D = X_train.shape
M = 512
batch_size = 300
epochs = 50
##### keras #####
from keras.models import Sequential
from keras.layers import Dense
# input: N,D W1: D,M W2: M,class_num
model = Sequential()
model.add(Dense(units=M, input_shape=(D, ),
activation='relu')) # input_dim = D， units为output个数
model.add(Dense(units=class_num))
from Data import DataExtract
from sklearn.mixture import GaussianMixture
from Data.DataTransform import purity, DBI
X_train, _, Y_train, _ = DataExtract.load_minist_csv()
model = GaussianMixture(n_components=10)
model.fit(X_train)
# VxD
M = model.means_
var = model.covariances_
# 根据数据预测各组分的后验概率。
# NxV
R = model.predict_proba(X_train)
# 分类后的
print("Purity:", purity(Y_train, R))
# 2个聚类间std偏差和/聚类均值间距离的比值 越低越好
print("DBI:", DBI(X_train, M, R))