ENGLISH #1.AC_process1.py:
-
main script of program。
-
if you are from the very begining of the experiment, please run following with main function. 1. use data_process() to generate data list. 2. use pretrain() to initialize model. 3. run 4 active learning method. 4. after all done, run Test_model_process() to generate confusion matrix and other value.
-
Experiment_NAME is quite important. If you already generated data and initialized model, step 1 and 2 could be ignored and you can run active learning directly.
#2.file:
-
performance index report is generate in name :Experiment_NAME + method_name + report, saved under confusion_matrix directory. txt file
-
precision,recall,f1-score generated in Test_model_process(). txt file
-
simple confusion matrix is generated in name:Experiment_NAME + method_name + stage, saved under confusion_matrix directory. black-white color. png file
-
confusion matrix is generated in name: Experiment_NAME + method_name + stage saved under confusion_matrix directory. Matrix. npy file
-
simple log is generated in name: Experiment_NAME + method_name, saved under simple_log directory. Simply saved precision in every stage. For bi-direction entropy selection, high confidence sample precision is also saved. txt file.
-
Plot_conf.py: plot confusion matrix
#3.Unit.py: load data
#4.layer.py: network layer function
#5.data format save image in 1,2,3....directory. Every directory contains 1 class cars image.
CHINESE
#1.AC_process1.py注意事项:
-
程序主脚本。
-
整个实验如果是从头开始,应当按照主函数流程进行。首先运行data_process()生成数据目录,然后运行pretrain()生成初始模型,然后运行四个主动学习过程,生成阶段模型所有模型生成完毕后,运行Test_model_process()来生成混淆矩阵以及各种相关值。
-
其中Experiment_NAME的设置很关键,如果你已经生成好了数据以及初始模型,则可以直接运行主动过程,而无需全部从新来过。
#2.生成文件一览:
-
指标report,命名规则:实验名+方法名+report,存放在confusion_matrix目录下。txt文件
-
执行Test_model_process()时生成的关于precision,recall,f1-score的报告,记录了每个阶段的每个类目的三项指标。txt文件
-
简易混淆矩阵,命名规则:实验名+方法名+阶段,存放在confusion_matrix目录下。颜色是黑白的。png文件
-
混淆矩阵,命名规则:实验名+方法名+阶段,存放在confusion_matrix目录下。数字形式,类目数乘类目数的一个数组矩阵。npy文件
-
简易日志,命名规则:实验名+方法,存放在simple_log文件夹下。简单记录了每一个阶段的准确率,对于双向算法同时记录了高置信度样本准确率。txt文件
-
Plot_conf.py 注意事项: 绘制混淆矩阵的脚本。载入数字混淆矩阵后绘图即可。
#3.Unit.py注意事项: 一些读取数据时的函数,具体看注释即可。
#4.layer.py注意事项: 网络层函数。
#5.数据格式 按1,2,3,4,5...的目录名存放车型图片,每个目录下面为一个类别的车辆。