# -*- coding: utf-8 -*-
# http://ju.outofmemory.cn/entry/344970 //layers
# https://my.oschina.net/u/876354/blog/1637819
# https://www.zhihu.com/question/49039504
import os
import tensorflow as tf
import 數(shù)據(jù)處理_hss as data_hss
import shutil
import numpy as np
# import 模型2sxc as md1
import time
def conv1d_relu(X,filters,kernel_size,strides):
X_change = tf.layers.conv1d(X, filters, kernel_size, strides, 'same', use_bias=True, activation=None)
X_change = tf.nn.relu(X_change)
return X_change
if __name__ == '__main__':
start = time.time()
file_1 = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\N' #文件夾
file_2 = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\S'
file_3 = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\T'
file_4 = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\V'
file_5 = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\X'
file_NO_X = r'D:\hss_python_work\resnet_cnn\數(shù)據(jù)集\非X'
data_all_1,label_all_1 = data_hss.data_analysis(file_NO_X,label_def = [0,1],proportion = 800) # 先定義標簽
data_all_2,label_all_2 = data_hss.data_analysis(file_5,label_def = [1,0],proportion = 50) # 先定義標簽
data_all = data_all_1 + data_all_2
label_all = label_all_1 + label_all_2
print("data_all = ",len(data_all))
print("label_all = ",len(label_all))
data_train,label_train,data_test,label_tast = data_hss.Dataset_partition(data_all,label_all)
print(data_train.shape,label_train.shape,data_test.shape,label_tast.shape)
sess = tf.InteractiveSession()
input_data = tf.placeholder(tf.float32, [None, 75])
input_label = tf.placeholder(tf.float32, [None, 2])
inputdata = tf.reshape(input_data,[-1,75,1])
conv1_7x7 = conv1d_relu(inputdata, 64, 7, 2)
pool1_3x3 = tf.layers.max_pooling1d(conv1_7x7, 3, 2, padding='same')
#深度學習的局部響應歸一化LRN層划乖,對局部神經(jīng)元的活動創(chuàng)建競爭機制聋袋,使得其中響應比較大的值變得相對更大均函,并抑制其他反饋較小的神經(jīng)元牙瓢,增強了模型的泛化能力擎析。
# pool1_3x3 = tf.nn.local_response_normalization(pool1_3x3) # 圖像
conv2_3x3_reduce = conv1d_relu(pool1_3x3, 64, 1, 1)
conv2_3x3 = conv1d_relu(conv2_3x3_reduce, 192, 3, 1)
# conv2_3x3 = tf.nn.local_response_normalization(conv2_3x3)
pool2_3x3 = tf.layers.max_pooling1d(conv2_3x3, 3, 2, padding='same')
# 3a ###########
inception_3a_1_1 = conv1d_relu(pool2_3x3, 64, 1, 1)
inception_3a_3_3_reduce = conv1d_relu(pool2_3x3, 96, 1, 1)
inception_3a_3_3 = conv1d_relu(inception_3a_3_3_reduce, 128, 3, 1)
inception_3a_5_5_reduce = conv1d_relu(pool2_3x3, 16, 1, 1)
inception_3a_5_5 = conv1d_relu(inception_3a_5_5_reduce, 32, 5, 1)
inception_3a_pool = tf.layers.max_pooling1d(pool2_3x3, 3, 1, padding='same')
inception_3a_pool_1_1 = conv1d_relu(inception_3a_pool, 32, 1, 1)
inception_3a_output = tf.concat([inception_3a_1_1,inception_3a_3_3,inception_3a_5_5,inception_3a_pool_1_1],axis=2)
# 3b ###########
inception_3b_1_1 = conv1d_relu(inception_3a_output, 128, 1, 1)
inception_3b_3_3_reduce = conv1d_relu(inception_3a_output, 128, 1, 1)
inception_3b_3_3 = conv1d_relu(inception_3b_3_3_reduce, 192, 3, 1)
inception_3b_5_5_reduce = conv1d_relu(inception_3a_output, 32, 1, 1)
inception_3b_5_5 = conv1d_relu(inception_3b_5_5_reduce, 96, 5, 1)
inception_3b_pool = tf.layers.max_pooling1d(inception_3a_output, 3, 1, padding='same')
inception_3b_pool_1_1 = conv1d_relu(inception_3b_pool, 64, 1, 1)
inception_3b_output = tf.concat([inception_3b_1_1, inception_3b_3_3, inception_3b_5_5,inception_3b_pool_1_1],axis=2)
pool3_3_3 = tf.layers.max_pooling1d(inception_3b_output, 3, 2, padding='same')
# 4a ###########
inception_4a_1_1 = conv1d_relu(pool3_3_3, 192, 1, 1)
inception_4a_3_3_reduce = conv1d_relu(pool3_3_3, 96, 1, 1)
inception_4a_3_3 = conv1d_relu(inception_4a_3_3_reduce, 208, 3, 1)
inception_4a_5_5_reduce = conv1d_relu(pool3_3_3, 16, 1, 1)
inception_4a_5_5 = conv1d_relu(inception_4a_5_5_reduce, 48, 5, 1)
inception_4a_pool = tf.layers.max_pooling1d(pool3_3_3, 3, 1, padding='same')
inception_4a_pool_1_1 = conv1d_relu(inception_4a_pool, 64, 1, 1)
inception_4a_output = tf.concat([inception_4a_1_1, inception_4a_3_3, inception_4a_5_5, inception_4a_pool_1_1],axis=2)
# 4b ###########
inception_4b_1_1 = conv1d_relu(inception_4a_output, 160, 1, 1)
inception_4b_3_3_reduce = conv1d_relu(inception_4a_output, 112, 1, 1)
inception_4b_3_3 = conv1d_relu(inception_4b_3_3_reduce, 224, 3, 1)
inception_4b_5_5_reduce = conv1d_relu(inception_4a_output, 24, 1, 1)
inception_4b_5_5 = conv1d_relu(inception_4b_5_5_reduce, 64, 5, 1)
inception_4b_pool = tf.layers.max_pooling1d(inception_4a_output, 3, 1, padding='same')
inception_4b_pool_1_1 = conv1d_relu(inception_4b_pool, 64, 1, 1)
inception_4b_output = tf.concat([inception_4b_1_1, inception_4b_3_3, inception_4b_5_5,inception_4b_pool_1_1],axis=2)
# 4c ###########
inception_4c_1_1 = conv1d_relu(inception_4b_output, 128, 1, 1)
inception_4c_3_3_reduce = conv1d_relu(inception_4b_output, 128, 1, 1)
inception_4c_3_3 = conv1d_relu(inception_4c_3_3_reduce, 256, 3, 1)
inception_4c_5_5_reduce = conv1d_relu(inception_4b_output, 24, 1, 1)
inception_4c_5_5 = conv1d_relu(inception_4c_5_5_reduce, 64, 5, 1)
inception_4c_pool = tf.layers.max_pooling1d(inception_4b_output, 3, 1, padding='same')
inception_4c_pool_1_1 = conv1d_relu(inception_4c_pool, 64, 1, 1)
inception_4c_output = tf.concat([inception_4c_1_1, inception_4c_3_3, inception_4c_5_5, inception_4c_pool_1_1],axis=2)
# 4d ###########
inception_4d_1_1 = conv1d_relu(inception_4c_output, 112, 1, 1)
inception_4d_3_3_reduce = conv1d_relu(inception_4c_output, 144, 1, 1)
inception_4d_3_3 = conv1d_relu(inception_4d_3_3_reduce, 288, 3, 1)
inception_4d_5_5_reduce = conv1d_relu(inception_4c_output, 32, 1, 1)
inception_4d_5_5 = conv1d_relu(inception_4d_5_5_reduce, 64, 5, 1)
inception_4d_pool = tf.layers.max_pooling1d(inception_4c_output, 3, 1, padding='same')
inception_4d_pool_1_1 = conv1d_relu(inception_4d_pool, 64, 1, 1)
inception_4d_output = tf.concat([inception_4d_1_1, inception_4d_3_3, inception_4d_5_5, inception_4d_pool_1_1],axis=2)
# 4e ###########
inception_4e_1_1 = conv1d_relu(inception_4d_output, 256, 1, 1)
inception_4e_3_3_reduce = conv1d_relu(inception_4d_output, 160, 1, 1)
inception_4e_3_3 = conv1d_relu(inception_4e_3_3_reduce, 320, 3, 1)
inception_4e_5_5_reduce = conv1d_relu(inception_4d_output, 32, 1, 1)
inception_4e_5_5 = conv1d_relu(inception_4e_5_5_reduce, 128, 5, 1)
inception_4e_pool = tf.layers.max_pooling1d(inception_4d_output, 3, 1, padding='same')
# 4e ###########
inception_4e_pool_1_1 = conv1d_relu(inception_4e_pool, 128, 1, 1)
inception_4e_output = tf.concat([inception_4e_1_1, inception_4e_3_3, inception_4e_5_5,inception_4e_pool_1_1],axis=2)
pool4_3_3 = tf.layers.max_pooling1d(inception_4e_output, 3, 2, padding='same')
# 5a ###########
inception_5a_1_1 = conv1d_relu(pool4_3_3, 256, 1, 1)
inception_5a_3_3_reduce = conv1d_relu(pool4_3_3, 160, 1, 1)
inception_5a_3_3 = conv1d_relu(inception_5a_3_3_reduce, 320, 3, 1)
inception_5a_5_5_reduce = conv1d_relu(pool4_3_3, 32, 1, 1)
inception_5a_5_5 = conv1d_relu(inception_5a_5_5_reduce, 128, 5, 1)
inception_5a_pool = tf.layers.max_pooling1d(pool4_3_3, 3, 1, padding='same')
inception_5a_pool_1_1 = conv1d_relu(inception_5a_pool, 128, 1, 1)
inception_5a_output = tf.concat([inception_5a_1_1, inception_5a_3_3, inception_5a_5_5, inception_5a_pool_1_1],axis=2)
# 5b ###########
inception_5b_1_1 = conv1d_relu(inception_5a_output, 384, 1, 1)
inception_5b_3_3_reduce = conv1d_relu(inception_5a_output, 192, 1, 1)
inception_5b_3_3 = conv1d_relu(inception_5b_3_3_reduce, 384, 3, 1)
inception_5b_5_5_reduce = conv1d_relu(inception_5a_output, 48, 1, 1)
inception_5b_5_5 = conv1d_relu(inception_5b_5_5_reduce, 128, 5, 1)
inception_5b_pool = tf.layers.max_pooling1d(inception_5a_output, 3, 1, padding='same')
inception_5b_pool_1_1 = conv1d_relu(inception_5b_pool, 128, 1, 1)
inception_5b_output = tf.concat([inception_5b_1_1, inception_5b_3_3, inception_5b_5_5, inception_5b_pool_1_1],axis=2)
##########
keep_prob = tf.placeholder(tf.float32)
pool5_7_7 = tf.layers.average_pooling1d(inception_3a_output, 3, 1, padding='valid') # inception_5b_output
pool5_7_7 = tf.layers.dropout(pool5_7_7, keep_prob)
print("pool5_7_7 = ",pool5_7_7.shape)
# 鋪平圖像數(shù)據(jù)
pool5_7_7_flat = tf.layers.Flatten()(pool5_7_7)
print("pool5_7_7_flat = ",pool5_7_7_flat.shape)
#全連接層計算
y_conv = tf.layers.dense(pool5_7_7_flat,2) #,activation = tf.nn.softmax
# print("y_conv = ",y_conv.shape)
# 計算交叉熵損失
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=input_label))
# 創(chuàng)建優(yōu)化器,通知Tensorflow在訓練時要更新均值和方差的分布
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
train_step = tf.train.AdamOptimizer(1e-2).minimize(cross_entropy)
# 創(chuàng)建計算準確度的操作
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(input_label, 1))
accuracy_rate = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
########開始訓練過程########
# 初始化所有變量、參數(shù)
tf.global_variables_initializer().run()
m_saver = tf.train.Saver(var_list=tf.global_variables(),max_to_keep = 15) #聲明tf.train.Saver類用于保存模型
########### 訓練(設置訓練時Dropout的kepp_prob比率為0.5逼泣。mini-batch為50辅柴,進行2000次迭代訓練,參與訓練樣本5萬)
model_doc = r'model' # 存儲模型的文件夾
if not os.path.exists(model_doc): # 模型存儲文件夾
os.makedirs(model_doc)
else:
shutil.rmtree(model_doc) #先強制刪除文件夾
os.makedirs(model_doc)
train_accuracy_all = [] #保存所有準確度
max_acc = 0
f = open('model/acc.txt','w')
for i in range(10000):
batch_data_train,batch_label_train = data_hss.batch(data_train,label_train,batch_size = 200)
# print("batch_data_train = ",batch_data_train,batch_data_train.shape)
# print("batch_label_train = ",batch_label_train,batch_label_train.shape)
# 循環(huán)次數(shù)是100的倍數(shù)的時候怠噪,打印
if i%1 == 0:
train_accuracy = accuracy_rate.eval(feed_dict={input_data:batch_data_train,input_label: batch_label_train,
keep_prob: 0.5})
print ("-->step %d, training accuracy %.4f ,max_acc %.4f"%(i, train_accuracy,max_acc))
print("cross_entropy = ",sess.run(cross_entropy,feed_dict={input_data:batch_data_train,input_label: batch_label_train,
keep_prob: 0.5}))
f.write(str(i)+', train_accuracy: '+str(train_accuracy)+' '+str(max_acc) +'\n')
# #保存最近的5次模型
# m_saver.save(sess, './model/model.ckpt', global_step=i)
#保存準確度最高的5次模型
if train_accuracy >= max_acc :
max_acc = train_accuracy
if train_accuracy >= max_acc - 0.04:
m_saver.save(sess, './model/model.ckpt', global_step=i)
train_accuracy_all.append(train_accuracy)
# 執(zhí)行訓練模型
train_step_,loss = sess.run([train_step,cross_entropy], feed_dict={input_data:batch_data_train,
input_label: batch_label_train,
keep_prob: 0.5})
# 打印測試集正確率 恐似,全部訓練完成之后,在最終測試集上進行全面測試傍念,得到整體的分類準確率
# print(train_accuracy_all)
if max_acc >= 0.9 and train_accuracy < 0.5: #訓練準確度過了最高點并下降矫夷,就停止訓練
break
f.close()
#######模型讀回及預測
model_file=tf.train.latest_checkpoint('model/')
m_saver.restore(sess,model_file)
#寫文件
# output1 = "偽差.txt"
# if os.path.exists(output1): os.remove(output1)
# file_name1 = open(output1, 'a+')
# output2 = "非偽差.txt"
# if os.path.exists(output2): os.remove(output2)
# file_name2 = open(output2, 'a+')
# output3 = "R_left2right.txt"
# if os.path.exists(output3): os.remove(output3)
# file_name3 = open(output3, 'a+')
# output4 = "R_left2other.txt"
# if os.path.exists(output4): os.remove(output4)
# file_name4 = open(output4, 'a+')
a = 0 # 預測對的個數(shù)
TP = 0 # 預測正確的病癥個數(shù)
FN_TP = 0 # 原標簽中有病癥的個數(shù)
TN = 0 # 預測正確的非病癥個數(shù)
TN_FP = 0 # 原標簽中正常的個數(shù)
sensibility = 0 # 敏感性
specificity = 0 # 特異性
single_data = np.empty([1,75])
single_label = np.zeros([1,2],"int")
for i in range(0,len(data_test)):
single_data[0] = data_test[i]
single_label = label_tast[i]
# print("single_data = ",single_data)
# print("single_label = ",single_label)
output = sess.run(y_conv , feed_dict = {input_data:single_data, keep_prob: 1.0})
# print("output = ",output)
output = sess.run(tf.nn.softmax(output))
output = np.round(output) #np.round(output) #取最接近的整數(shù)
print(i,"/",len(data_test)-1," output = ",output,"single_label = ",single_label)
if single_label[0] == output[0][0] and single_label[1] == output[0][1] :
a +=1
if single_label[0] == output[0][0] and output[0][0] == 1: #敏感性
TP += 1
if single_label[0] == 1 :
FN_TP += 1
if single_label[1] == output[0][1] and output[0][1] == 1: #特異性
TN += 1
if single_label[1] == 1 :
TN_FP += 1
# if output[0][0] == 1: # 偽差
# file_name1.write(str(data_test[i]) + '\n')
# if output[0][1] == 1: # 非偽差
# file_name2.write(str(data_test[i]) + '\n')
# if single_label[0] == 1 and output[0][0] == 0: # 偽差2非偽差
# file_name3.write(str(data_test[i]) + '\n')
# if single_label[1] == 1 and output[0][0] == 1: # 非偽差2偽差
# file_name4.write(str(data_test[i]) + '\n')
print("len(data_test) = ",len(data_test),"a =",a)
print("sensibility = ",TP/FN_TP,"specificity =",TN/TN_FP)
# file_name1.close()
# file_name2.close()
# file_name3.close()
# file_name4.close()
end = time.time()
print("程序運行時間:",end - start)
inception_v1網(wǎng)絡
最后編輯于 :
?著作權歸作者所有,轉載或內(nèi)容合作請聯(lián)系作者
- 文/潘曉璐 我一進店門,熙熙樓的掌柜王于貴愁眉苦臉地迎上來评矩,“玉大人叶堆,你說我怎么就攤上這事〕舛牛” “怎么了虱颗?”我有些...
- 文/不壞的土叔 我叫張陵,是天一觀的道長蔗喂。 經(jīng)常有香客問我上枕,道長,這世上最難降的妖魔是什么弱恒? 我笑而不...
- 正文 為了忘掉前任,我火速辦了婚禮棋恼,結果婚禮上返弹,老公的妹妹穿的比我還像新娘锈玉。我一直安慰自己,他們只是感情好义起,可當我...
- 文/花漫 我一把揭開白布拉背。 她就那樣靜靜地躺著,像睡著了一般默终。 火紅的嫁衣襯著肌膚如雪椅棺。 梳的紋絲不亂的頭發(fā)上,一...
- 文/蒼蘭香墨 我猛地睜開眼勺美,長吁一口氣:“原來是場噩夢啊……” “哼!你這毒婦竟也來了碑韵?” 一聲冷哼從身側響起赡茸,我...
- 正文 年R本政府宣布蜈抓,位于F島的核電站启绰,受9級特大地震影響,放射性物質發(fā)生泄漏沟使。R本人自食惡果不足惜委可,卻給世界環(huán)境...
- 文/蒙蒙 一、第九天 我趴在偏房一處隱蔽的房頂上張望。 院中可真熱鬧着倾,春花似錦拾酝、人聲如沸。這莊子的主人今日做“春日...
- 文/蒼蘭香墨 我抬頭看了看天上的太陽。三九已至崇决,卻和暖如春材诽,著一層夾襖步出監(jiān)牢的瞬間,已是汗流浹背恒傻。 一陣腳步聲響...
推薦閱讀更多精彩內(nèi)容
- 原文鏈接:blog.csdn.net/langb2014/article/details/52787095 來自c...
- Http協(xié)議庫 HttpUrlConnetion同步處理網(wǎng)絡請求底層支持Http1.0,Http1.1栅隐。缺點:An...
- 簡單示例 屬性歸類 事件簡介 相關文檔 https://nodejs.org/api/net.html#net_s...
- 論文分享 蘿卜兔編輯整理 卷積神經(jīng)網(wǎng)絡廣泛應用在各種任務塔嬉,比如圖像分類、目標檢測等租悄,性能也越來越好谨究,但都趨向于使...
- SiameseFC: Luca Bertinetto, Jack Valmadre, Jo?o F. Henriq...