Kaggle "Microsoft Malware Classification Challenge"——就是沙箱恶意文件识别，有 Opcode n-gram特征 ASM文件图像纹理特征 还有基于图聚类方法
使用图聚类方法：Malware Classification using Graph Clustering 见 https://github.com/rahulp0491/Malware-Classifier
代码参考：https://github.com/bindog/ToyMalwareClassification，https://github.com/xiaozhouwang/kaggle_Microsoft_Malware
#微软恶意代码分类
比赛说明和数据下载 https://www.kaggle.com/c/malware-classification/
##代码说明
randomsubset.py抽取训练子集asmimage.pyASM文件图像纹理特征opcode_n-gram.pyOpcode n-gram特征firstrandomforest.py基于ASM文件图像纹理特征的随机森林secondrandomforest.py基于Opcode n-gram特征特征的随机森林combine.py将两种类型的特征结合
##运行说明
- 将完整的训练数据集解压，修改
randomsubset.py中的路径并运行 - 修改
asmimage.py和opcode_n-gram.py中的路径，并运行run.sh，耐心等待即可看到结果
参考:https://github.com/dchad/malware-detection
malware-detection
Experiments in malware detection and classification using machine learning techniques.
1. Microsoft Malware Classification Challenge
https://www.kaggle.com/c/malware-classification
1.1 Feature Engineering
Initial feature engineering consisted of extracting various keyword counts from the ASM files
as well as the entropy and file size from the BYTE files of the 10868 malware samples in the training set.
Image files of the first 1000 bytes of the ASM and BYTE files were created and combined with
keyword and entropy data. This resulted in a set of 2018 features.
Flow control graphs and call graphs were generated for each ASM sample. A feature set was
then generated from the graphs, including graph maximum delta, density, diameter and function
counts etc.
1.2 Feature Selection
Statistical analysis of the feature set using chi-squared tests to remove features that are
independent of the class labels or have low variance. The BYTE file images were found to be weak
learners and were removed from the feature set. A comparison of the best features from the chi-squared
tests with reduced feature sets of between 10% - 50% of the original features.
1.2.1 Selection Comparison
Testing with an ExtraTreesClassifier and 10-fold cross validation produced the following results:
- Original ASM Keyword Counts (1006 features): logloss = 0.034
- 10% Best ASM Features with Entropy and Image Features (202 features): logloss = 0.0174
- 20% Best ASM with Entropy and Image Features (402 features): logloss = 0.0164
- 30% Best ASM with Entropy and Image Features plus Feature Statistics (623 features):
multiclass logloss = 0.0133
accuracy score = 0.9978
Confusion Matrix:
[[1540 0 0 0 0 1 0 0 0]
[ 1 2475 2 0 0 0 0 0 0]
[ 0 0 2942 0 0 0 0 0 0]
[ 1 0 0 474 0 0 0 0 0]
[ 2 0 0 0 38 2 0 0 0]
[ 3 0 0 0 0 748 0 0 0]
[ 1 0 0 0 0 0 397 0 0]
[ 0 0 0 0 0 0 0 1225 3]
[ 0 0 0 0 0 0 0 8 1005]]
- 40% Best ASM and image features with feature statistics:
ExtraTreesClassifier with 1000 estimators on 10868 training samples and 823 features
using 10-fold cross validation:
multiclass logloss = 0.0135
accuracy score = 0.9976
Confustion Matrix:
[[1541 0 0 0 0 0 0 0 0]
[ 1 2475 2 0 0 0 0 0 0]
[ 0 0 2942 0 0 0 0 0 0]
[ 1 0 0 474 0 0 0 0 0]
[ 5 0 0 0 37 0 0 0 0]
[ 5 0 0 0 0 746 0 0 0]
[ 1 0 0 0 0 0 397 0 0]
[ 0 0 0 0 0 0 0 1227 1]
[ 0 0 0 0 0 0 0 9 1004]]
1.2.2 Feature Selection Summary
The performance of the ExtraTreesClassifier is optimal at around 30% of ASM and image features
with highest variance plus sample statistics, entropy and file size. Adding call graph features
produced a marginal improvement. It is possible that better classification accuracy would be
achieved by using an ensemble of different classifiers with the ASM, image and call graph
feature sets as separate inputs to the various classifiers.
1.3 Model Selection
Selection of candidate models using GridSearchCV to find optimal classifier hyper-parameters.
- SVM:
- ExtraTrees:
- XGBoost: 30% Best Features
logloss: 0.0080
accuracy: 0.9981
Confusion Matrix:
[[1540 0 0 0 0 1 0 0 0]
[ 2 2475 0 1 0 0 0 0 0]
[ 0 0 2941 0 0 0 1 0 0]
[ 0 0 0 474 0 1 0 0 0]
[ 1 0 0 0 41 0 0 0 0]
[ 4 0 0 0 1 746 0 0 0]
[ 0 0 0 0 0 0 398 0 0]
[ 0 0 0 0 0 0 0 1227 1]
[ 0 0 0 0 0 0 0 8 1005]]
- NaiveBayes:
- KNN:
1.4 Graphs
1. Shannon's Entropy by malware class. A score of 0.0 means the bytes are all the same value, a score of 1.0 means every byte in the file has a different value.
2. Shannon's Entropy by file size. A score of 0.0 means the bytes are all the same value, a score of 1.0 means every byte in the file has a different value.
3. Assembler register EDX by ESI counts.
1.5 Conclusions
The best accuracy scores were achieved with XGBoost (99.81%) and ExtraTreesClassifier (99.76%) using a feature set of 623 ASM, image and entropy features. Marginal improvements could be achieved using additional features and ensemble methods, however due to the limited sample size further efforts are unlikely to produce significant improvements in prediction accuracy. Analysis will now focus on much larger sample sizes from VirusShare.com as described in the following sections.
<<<=============================================================>>>
2. VirusShare.com Malware Collection Analysis
VirusShare.com regularly publishes huge collections of malware binaries for use by researchers. Each malware archive is currently around 25GB in size. Several of the latest archives have been downloaded to use as training and test sets. The archives used are:
- Training set: VirusShare_00251.zip and VirusShare_00252.zip (131072 malware samples)
VirusShare_00263.zip and VirusShare_00264.zip (131072 malware samples)
VirusShare_APT1_293.zip (293 malware samples)
- Testing set:
2.1 Automated Unpacking and Disassembly of Malware Binaries
Using Cuckoo Sandbox and unpack.py for behaviourial analysis, unpacking the binaries and
dumping process memory, for intransigent samples, manual unpacking with Immunity Debugger and IDA Pro.
Tools:
- Cuckoo Sandbox (https://github.com/cuckoosandbox/cuckoo)
- unpack.py (https://malwaremusings.com/2013/02/26/automated-unpacking-a-behaviour-based-approach/)
(https://github.com/malwaremusings/unpacker)
- IDA Pro 5.0 (https://www.hex-rays.com/products/ida/support/download_freeware.shtml)
- Immunity Debugger (https://www.immunityinc.com/products/debugger/)
- Volatility (https://github.com/volatilityfoundation)
- Ildasm.exe (https://msdn.microsoft.com/en-us/library/f7dy01k1(v=vs.110).aspx)
- ndisasm (http://www.nasm.us/pub/nasm/releasebuilds/2.12.02/)
- TrID (http://mark0.net/soft-trid-e.html)
- ClamAV (clamav.net)
- Windows Defender
- MalwareBytes Anti-Malware
- VirusTotal.com
Environment Setup (Debian):
apt install virtualbox virtualbox-dkms python-dev libffi-dev virtualenv virtualenvwrapper clamav
pip install cython numpy scipy scikit-learn matplotlib jupyter pandas xgboost
git clone https://github.com/cuckoosandbox/cuckoo
git clone https://github.com/volatility
Environment Setup (Windows):
TODO:
2.2 Generating Training Labels
ClamAV and Windows Defender used for initial training label generation or VirusTotal.com aggregate classification
if they cannot identify the culprit. MalwareBytes was also used but it crashed at the end of the scan
and the log files could not be recovered.
AV Scan Results:
Results: VirusShare_00251
- 57529 files classified as malicious.
- 8007 files classified as non-malicious.
Results: VirusShare_00252
- 56625 files classified as malicious.
- 8911 files classified as non-malicious.
Results: VirusShare_00263
- 51612 files classified as malicious.
- 13924 files classified as non-malicious.
Results: VirusShare_00264
- 42274 files classified as malicious.
- 23262 files classified as non-malicious.
Results: VirusShare_APT1_293
- 292 files classified as malicious.
- 1 file classified as non-malicious.
Total Malware Types: 8334
Total Malware Families: 2737
Total Files: 262437
2.2.1 Graphs
4. Top 10 Malware Counts.
5. Top 10 Compiler/Packer Counts.
6. VirusShare 251 Call Graph - Vertex by Edge Count.
7. VirusShare 251 Shannon's File Entropy Histogram.
2.3 Converting to ASM and Feature Extraction
IDA Pro and objdump for disassembly of binaries to .asm text files.
Feature sets will consist of:
- Entropy and file size from packed binaries.
- Entropy and file size from unpacked binaries.
- File magic signatures and TrID signatures.
- ASM features from disassembled unpacked binaries.
- Executable header features.
- Call Graph Features.
- Function counts extracted from call graphs.
- Sample Statistics.
- Behavioural features from Cuckoo Sandbox reports.
- Memory features from Volatility reports.
2.4 Feature Selection and Reduction
1. PE/COFF Binaries: (Chi2 Tests)
VS251 Feature Sets: 54911 samples.
240 PE ASM and Header Features.
?? PE ASM Function Count Features.
VS252 Feature Sets: 46165 samples.
271 PE ASM and Header Features.
?? PE ASM Function Count Features.
VS263 Feature Sets: 40974 samples.
203 PE ASM and Header Features.
?? PE ASM Function Count Features.
VS264 Feature Sets: 14366 samples.
243 PE ASM and Header Features.
?? PE ASM Function Count Features.
2. ELF Binaries:
3. Java Bytecode:
4. Javascript:
5. HTML:
6. PDF:
2.5 Model Selection
2.5.1 PE/COFF Model Selection
Model selection with 10-fold cross validation:
1. ExtraTreesClassifier: VS251 100 estimators accuracy score = 0.912
500 estimators accuracy score = ?.??
1000 estimators accuracy score = memory fail
VS252 100 estimators accuracy score = 0.888 (12.75 minutes)
500 estimators accuracy score = ?.??
1000 estimators accuracy score = ?.??
VS263 100 estimators accuracy score = 0.903 (9.63 minutes)
500 estimators accuracy score = ?.???
1000 estimators accuracy score = ?.??
VS264 100 estimators accuracy score = 0.889 (2.27 minutes)
500 estimators accuracy score = 0.890 (14.57 minutes)
1000 estimators accuracy score = ?.??
2. XGBoost: VS251 100 estimators accuracy score = ?.??
XGBoost: VS252 100 estimators accuracy score = ?.??
XGBoost: VS263 100 estimators accuracy score = ?.??
XGBoost: VS264 100 estimators accuracy score = ?.??
3. LightGBM: VS251 100 estimators accuracy score = 0.892
VS252 100 estimators accuracy score = 0.676 (171.23 minutes)
VS263 100 estimators accuracy score = ?.??
VS264 100 estimators accuracy score = 0.758 (9.26 minutes)
200 estimators accuracy score = 0.750 (18.53 minutes)
4. RandomForestClassifier: VS251 100 estimators accuracy score = 0.903
500 estimators accuracy score = ?.??
1000 estimators accuracy score = ?.??
VS252 100 estimators accuracy score = 0.881 (81.34 minutes)
VS263 100 estimators accuracy score = ?.??
VS264 100 estimators accuracy score = 0.879 (15.45 minutes)
Model Stacks/Ensembles:
1. One input layer of classifiers -> 1 output layer classifier.
Layer 1: Six x layer one classifiers: (ExtraTrees x 2, RandomForest x 2, XGBoost x 1, LightGBM x 1)
Layer 2: One classifier: (ExtraTrees) -> final labels
2. Voting (Democratic and weighted).
Democratic: Six x layer one classifiers: (ExtraTrees x 2/RandomForest x 2/XGBoost/LightGBM)
-> (democratic vote, geometric and sum means) -> final labels
Weighted: Six x layer one classifiers: (ExtraTrees x 2/RandomForest x 2/XGBoost/LightGBM)
-> (weighted vote: ExtraTrees double weight, geometric and sum means) -> final labels
3. Multiple layers of classifiers.
Layer one -> layer two -> layer 3 -> final labels:
Layer 1: ExtraTrees x 2, RandomForest x 2, XGBoost x 1, LightGBM x 1
Layer 2: ExtraTrees x 2, RandomForest x 2, XGBoost x 1, LightGBM x 1
Layer 3: ExtraTrees x 1
4. Combined PE/COFF features + function count features:
Layer 1 -> layer 2 -> final labels
Layer 1 (A MODELS): Combined features layer one (ExtraTrees x 2, RandomForest x 2, XGBoost x 1, LightGBM x 1)
Layer 1 (B MODELS): Function count features layer one (ExtraTrees x 2, RandomForest x 2, XGBoost x 1, LightGBM x 1)
Layer 2: ExtraTrees x 1 -> final labels
5. Combine outputs from 1, 2, 3 and 4 -> vote -> final labels
2.5.2 ELF Model Selection
2.5.3 Java Bytecode Model Selection
2.5.4 Javascript Model Selection
TODO:
2.6 Conclusions
TODO:
2.7 Workflows
2.7.1 Training Label Generation
1. Antivirus scans using ClamAV and Windows Defender.
> clamscan -v -r /directory/containing/the/nastiness > clamav-report.txt
> Windows Defender (See notes in section 7 on extracting windows defender logs).
2. Generate scalar training labels for each malware type and family.
> process_av_reports.py
> combine_av_reports.py
> generate_train_labels.py
2.7.2 Feature Engineering
2.7.2.1 PE/COFF Malware Features
1. File entropy feature generation.
> feature_extraction_entropy.py
2. File magic signature and TrID signature feature generation.
> trid_check_file.py
> generate_file_ids.py
> feature_extraction_file_id.py
3. Packer identification feature generation.
> generate_packer_ids.py
> feature_extraction_packer_id.py
4. ASM feature generation (unpacked PE files).
> disassemble_pe.py
> feature_extraction_pe_asm.py
> generate_pe_header_tokens.py
> feature_extraction_pe_header.py
5. ASM feature generation (packed PE files).
> TODO:
6. Call Graph Generation and feature extraction.
> generate_call_graphs_pe_asm.py
> generate_function_column_names.py
> function_name_clean.py
> feature_extraction_pe_function_counts.py
> feature_reduction_pe_function_counts.py
7. Behavioural analysis feature generation.
> TODO:
8. Memory analysis feature generation.
> TODO:
2.7.2.2 ELF Malware Features
1. File entropy feature generation.
> feature_extraction_entropy.py
2. File magic signature and TrID signature feature generation.
> trid_check_file.py
> generate_file_ids.py
> feature_extraction_file_id.py
3. Packer identification feature generation.
> generate_packer_ids.py
> feature_extraction_packer_id.py
4. ASM feature generation.
> disassemble_elf.py
> feature_extraction_elf_asm.py
5. Call Graph Generation.
>
6. Behavioural analysis feature generation.
>
7. Memory analysis feature generation.
>
2.7.2.3 Java Bytecode Features
1. Convert Bytecode to Tokens.
2. Extract Bytecode Features.
Tools:
javap (https://docs.oracle.com/javase/7/docs/technotes/tools/windows/javap.html)
2.7.2.4 Javascript/HTML Features
1. Generate Javascript/HTML Keywords.
2. Unpack Javascript.
3. Extract Javascript/HTML Features.
Tools:
2.7.2.5 PDF Features
1. Generate PDF Keywords.
2. Extract Javascript/Shellcode/Macros.
3. Extract PDF Feature Sets.
Tools:
peepdf (https://github.com/jesparza/peepdf)
2.7.3 Feature Selection
2.7.3.1 PE/COFF Feature Selection
1. PE/COFF Feature Reduction.
> feature_reduction_pe_asm.py
> feature_reduction_pe_header.py
> feature_reduction_pe_function_counts.py
2.74 Model Selection
TODO:
1. PE/COFF Model Selection.
> model_selection_pe_coff.py
<<<========================================================>>>
3. Automated Sensor Malware Detection
TODO:
4. References
TODO:
5. Notes on installing xgboost for Python.
5.1 Source Install.
If installing from source, after building and installing you have problems loading other packages it is because of the xgboost-0.4-py2.7.egg.pth file that the install script dumps in the python dist-packages directory. You will have to delete the .pth file then go change the installation of the xgboost egg and egg-info files in the python dist-packages directory from:
/usr/local/lib/python2.7/dist-packages/xgboost-0.4-py2.7.egg/EGG_INFO
to:
/usr/local/lib/python2.7/dist-packages/xgboost-0.4-py2.7.dist-info
and:
/usr/local/lib/python2.7/dist-packages/xgboost-0.4-py2.7.egg/xgboost
to:
/usr/local/lib/python2.7/dist-packages/xgboost
Now python will be able to find all the packages.
5.2 Pip Install.
pip install xgboost
Now works for version 0.6a2 on Debian/Ubuntu/Mint distros.
5.3 Anaconda Install.
XGBoost is not a part of the official distribution but several community members have created Conda packages for it. The most up to date package seems to be by user creditx. The following command will install the package:
conda install -c creditx xgboost
6. Notes on Installing Cuckoo Sandbox
Python 2.7 is preferred for Cuckoo Sandbox, attempting with Python 3.x will be a fail. Installing the Python module requirements in requirements.txt results in failure because the module dpkt is only compatible with Python 2.x versions. If using Anaconda or python 3.x then revert to Python 2.7 or use mkvirtualenv to create a virtual environment to run cuckoo.
For example:
mkvirtualenv -p /usr/bin/python cuckoosandbox
Note: If using Anaconda: Remove the Anaconda bin directory from $PATH or it will cause an error when setting up the virtual environment. Also ensure that libxml2-dev and libxslt1-dev are installed or there will be build errors when installing the requirements.
7. Notes on Extracting Windows Defender Logs
Open a powershell - (Run as Administrator).
Enter the following commands:
> cd \Program Files\Windows Defender
> .\MpCmdRun -getfiles -scan
Several .log and .cab files will be placed in:
C:\ProgramData\Microsoft\Windows Defender\Support\
The Windows Defender malware detection log is called MPDetection-yymmdd-hhmm.log
8. Notes on Multi-Architecture Disassembly with objdump
Ensure binutils multi-target support has been installed (Linux Mint 18):
(NOTE: Linux Mint 17 does not have MIPS architecture in binutils, have to install from sauce.)
apt install binutils binutils-aarch64-linux-gnu binutils-alpha-linux-gnu binutils-arm-linux-gnueabi
binutils-arm-linux-gnueabihf binutils-arm-linux-gnueabihf binutils-arm-none-eabi binutils-avr binutils-dev
binutils-doc binutils-gold binutils-h8300-hms binutils-hppa-linux-gnu binutils-hppa64 binutils-hppa64-linux-gnu
binutils-m68hc1x binutils-m68k-linux-gnu binutils-mingw-w64 binutils-mingw-w64-i686 binutils-mingw-w64-x86-64
binutils-mips-linux-gnu binutils-mips64-linux-gnuabi64 binutils-mips64-linux-gnuabi64 binutils-mips64el-linux-gnuabi64
binutils-mips64el-linux-gnuabi64 binutils-mipsel-linux-gnu binutils-msp430 binutils-multiarch binutils-multiarch-dev
binutils-powerpc-linux-gnu binutils-powerpc-linux-gnuspe binutils-powerpc-linux-gnuspe binutils-powerpc64-linux-gnu
binutils-powerpc64-linux-gnu binutils-powerpc64le-linux-gnu binutils-powerpc64le-linux-gnu binutils-s390x-linux-gnu
binutils-sh4-linux-gnu binutils-source binutils-sparc64-linux-gnu binutils-z80 elf-binutils
9. Notes on Installing LightGBM
1. Clone, build and install:
git clone --recursive https://github.com/Microsoft/LightGBM
cd LightGBM
mkdir build
cd build
cmake ..
make -j
cd ../python-package/
python setup.py install
2. If you have problems with building or installing python module:
apt update
apt upgrade
apt install cmake
pip install setuptools numpy scipy scikit-learn -U
3. If you have problems with updating setuptools, sklearn etc, and you probably will because (pip == train wreck):
apt purge -y python-pip
wget https://bootstrap.pypa.io/get-pip.py
python ./get-pip.py
apt install python-pip
pip install setuptools numpy scipy scikit-learn -U
Kaggle "Microsoft Malware Classification Challenge"——就是沙箱恶意文件识别，有 Opcode n-gram特征 ASM文件图像纹理特征 还有基于图聚类方法的更多相关文章
- Kaggle ： Display Advertising Challenge( ctr 预估 )
原文:http://blog.csdn.net/hero_fantao/article/details/42747281 Display Advertising Challenge --------- ...
- cdmc2016数据挖掘竞赛题目Android Malware Classification
http://www.csmining.org/cdmc2016/ Data Mining Tasks Description Task 1: 2016 e-News categorisation F ...
- 转载一份kaggle的特征工程：经纬度、特征构造、转化率
转载:https://www.toutiao.com/i6642477603657613831/ 1 如果训练/测试都来自同一时间线,那么就可以非常巧妙地使用特性.虽然这只是一个kaggle的案例,但 ...
- 准确率99%！基于深度学习的二进制恶意样本检测——瀚思APT 沙箱恶意文件检测使用的是CNN，LSTM TODO
所以我们的流程如图所示.将正负样本按 1:1 的比例转换为图像.将 ImageNet 中训练好的图像分类模型作为迁移学习的输入.在 GPU 集群中进行训练.我们同时训练了标准模型和压缩模型,对应不同的 ...
- Pftriage：分析和追踪恶意文件，识别特征
项目地址 PFTriage:https://github.com/idiom/pftriage 参考 Pftriage:如何在恶意软件传播过程中对恶意文件进行分析 https://www.freebu ...
- 未能写入输出文件“c:\Windows\Microsoft.NET\Framework\v4.0.30319\Temporary ASP.NET Files\web\4b49f661\23a749fc\App_Web_default.aspx.cdcab7d2.zii776dc.dll”--“拒绝访问。 ” 解决方法:
解决方法: 找到C:\Windows\Temp 目录,在其属性->安全->编辑->添加 IIS_IUSERS用户 赋予"完全控制"权限 ...注意:这可能不是最佳 ...
- 中文版 Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks 摘要 最先进的目标检测网络依靠区域提出算法 ...
- 论文翻译：2018_Source localization using deep neural networks in a shallow water environment
论文地址:https://asa.scitation.org/doi/abs/10.1121/1.5036725 深度神经网络在浅水环境中的源定位 摘要: 深度神经网络(DNNs)在表征复杂的非线性关 ...
- 转载：LBP的初步理解
转自http://blog.csdn.net/ty101/article/details/8905394 本文的PDF版本,以及涉及到的所有文献和代码可以到下列地址下载: 1.PDF版本以及文献:ht ...
随机推荐
- Eigen库笔记整理（二）
Eigen/Geometry 模块提供了各种旋转和平移的表示 旋转矩阵直接使用 Matrix3d 或 Matrix3f Eigen::Matrix3d rotation_matrix = Eigen: ...
- list查询棚舍面积的时候,所有棚舍面积的value都是一样的
解决办法 将pickingid在查list之前set到对象中,通过id来匹配查询 具体代码: FarmHouse farmHouse=new FarmHouse(); farmHouse.setPic ...
- block的作用
ios高效开发--blocks相关 1.替换delegate 如果我们有2个viewController,a和b,当我们从a界面push到b后,在b上面触发了一些事件,这些时间又会影响 ...
- css 实现鼠标滑过流光效果
来划我啊 <!DOCTYPE html> <html> <head> <meta charset="UTF-8"> <titl ...
- C++ CEF 浏览器中显示 Tooltip(标签中的 title 属性)
在 Windows 中将 CEF 集成到 C++ 客户端以后,默认是无法显示 tooltip 的,比如图片标签中的 title 属性. 实现的方式其实很简单,按下面的步骤操作就可以: 创建一个文本文件 ...
- (C/C++学习)14.C语言字符串处理函数(二)
说明:上节着重解释了字符串处理的库函数处理,这节将针对一些常用的需求,进行非库函数的处理. 一.去除某一个字符串中的某个字符 1.去除字符串右边的空格 void trimStrRightSpace(c ...
- manacher（马拉车）算法
断断续续地看了两天的马拉车算法,可算是给搞明白了(贼开心),这算是自己搞懂的第一个算法了(23333333333333)这个算法照目前自己的理解来看,貌似就只能求个字符串中的回文串(接触这个算法是要求 ...
- Find The Multiple POJ - 1426 （BFS）
题目大意 给定一个整数,寻找一个只有0,1构成的十进制数使得这个数能够整除这个整数 解法 直接bfs第一位放入1,之后每一位放入1或者0 代码 #include <iostream> #i ...
- python爬虫29 | 使用scrapy爬取糗事百科的例子，告诉你它有多厉害！
是时候给你说说 爬虫框架了 使用框架来爬取数据 会节省我们更多时间 很快就能抓取到我们想要抓取的内容 框架集合了许多操作 比如请求,数据解析,存储等等 都可以由框架完成 有些小伙伴就要问了 你他妈的 ...
- 56-混沌操作法之我见：二、AO、AC指标.(2015.2.9)
混沌操作法之我见:二.AO.AC指标 先看看其算法: Y=(H+L)/2: AO=MA(Y,5)-MA(Y,34): AC=AO-MA(AO,5). 由算法可以看出,AO表示的是近5期的综合价格与近3 ...