# LanlExp

**Repository Path**: corgile/lanl-exp

## Basic Information

- **Project Name**: LanlExp
- **Description**: lanl-experiments
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2023-03-08
- **Last Updated**: 2023-03-10

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# LANL Experiments

This folder contains the code that ran the experiments we report on the LANL data set.

## Prerequisites

Please download the LANL data set from [https://csr.lanl.gov/data/cyber1/](https://csr.lanl.gov/data/cyber1/). Then,
in `./loaders/split.py` please specify the file locations, and output location for the cleaned, split data on lines
13-15 and run it. Finally, on line 17 of `./loaders/load_lanl.py` please specify the location of the LANL files.

To test the prior works, use the file `prior_works.py`. The command line flags `-v` (default), `-o`, and `-h` evaluate
VGRNN, EGCN_H-O, and EGCN_H-H respectively.

## Usage

To run experiments on the LANL data set, use `main.py`. Tests used in the paper are automated in `runall_LANL.sh`
and `run_delta.sh`. Options include:

    -h, --help    
        Show this help message and exit    
    -d DELTA, --delta DELTA                     (默认: 0.5 小时)
        快照时间跨度(hour), 默认0.5小时    
    -w WORKERS, --workers WORKERS               (默认: 8)
        工作进程数量（GNN个数）
    -T THREADS, --threads THREADS               (默认: 1)
        每个进程下的线程数
    -e {GCN,GAT,SAGE}, --encoder {GCN,GAT,SAGE} (默认: GCN)
        Worker进程要使用的GNN模型
    -r {GRU,LSTM,NONE}, --rnn {GRU,LSTM,NONE}   (默认: GRU)    
        Leader进程要使用的RNN模型    
    -H HIDDEN, --hidden HIDDEN                  (默认: 32)    
        隐藏层的维数（层数）
    -z ZDIM, --zdim ZDIM                        (默认: 16)     
        输出层的维度     
    -n NGRUS, --ngrus NGRUS                     (默认: 1)    
        Worker进程的RNN数量    
    -t TESTS, --tests TESTS                     (默认: 1)    
        需要进行几次独立测试    
    -l, --load                                  (默认: False)    
        如果使用此选项，直接加载上一次保存的模型，而不是重新训练
    --fpweight FPWEIGHT                         (默认: 0.6)    
        lambda 参数
    --nowrite    
        如果使用此选项，将不会将输出数据写入文件
    --impl {DETECT,PREDICT,D,P,PRED}, -i {DETECT,PREDICT,D,P,PRED}
        (默认: DETECT) 
        使用检测{DETECT,D} 还是预测{PREDICT,P,PRED}
    --dataset DATASET                           (默认: LANL)
        选择要使用的数据集（目前不可用）
    --lr LR                                     (默认: 0.005)    
        训练的学习率
    --patience PATIENCE                         (默认: 5)    
        PATIENCE个训练epoch后没有性能提升则直接结束训练（单位：佰）

Note: by default, this script uses ports 22032 and 22204 for RPC communication.