LogSummary

Requirements

Requirements are listed in requirements.txt.

gensim=3.8.1 numpy=1.17.2 networkx=2.4 rouge=1.0.0

To install these, run:

pip install -r requirements.txt

OpenIE Methods Installation

For most methods, it requires to have Java installed additionally to Python as it runs third party tools.

Stanford

Install the Stanford CoreNLP from here. Then update the openie.ini config file within the openie package.

Ollie

Install Ollie using their "Local Machine" installation process you can find here. Then update the openie.ini config file within the openie package.

OpenIE5

Install OpenIE5 using their pre-compiled stand-alone JAR you can find here.

Before using OpenIE5 you will need to run it as a server using a command similar to this one: java -Xmx10g -XX:+UseConcMarkSweepGC -jar openie-assembly-5.0-SNAPSHOT.jar --httpPort 8000 executing the downloaded stand-alone JAR. This is explained here in their repo.

Additionally, it requires a python wrapper you can find here which is already installed when you install the requirements.txt.

PredPatt

Proceed to install PredPatt as they explain it in their repo here.

ClausIE

Download ClauseIE from their source here. Update the openie.ini file using its directory for the jar location.

PropS

Clone [this repo]() where PropS has been upgraded to Python 3.8. Then you will need to update the config file at openie.ini and specify its package directory.

Configuration File

After installing the OpenIE methods above, make sure to update the openie.ini configuration file located inside the openie package according to your installation. It provides part of the settings for running the OpenIE methods that depend on Java such as StanfordNLP or external Python packages such as PropS.

Quick Start

Input Format

Templates

{
    "<id1>": [
        "< online template >",
        "< ground truth template >",
        [
            [
                "arg1",
                "predicate",
                "arg2"
            ],
            [< more triples >],
            [< more triples >]
        ]
    ],
    "<id2>":[...]
 }

Raw logs

Although logs have more freedom in their format as their preprocessing details are specific to each log type, the current expected format of most logs is the following.

<log_idx1>\t<log_message1>
<log_idx2>\t<log_message2>
...

Only the original switch logs format is expected to not have and index and be simply the log message.

<log_message1>
<log_message2>
...

Run LogIE

After the installation, run the following command in the home directory where this project is located.

python -m LogIE.run --templates "<Templates File>" --evaluation lexical --rules new --openie predpatt

Arguments

Runs information extraction from logs.

arguments:
  -h, --help            show this help message and exit
  --templates templates
                        input raw templates file path (default: None)
  --raw_logs raw_logs   input raw raw_logs file path (default: None)
  --base_dir base_dir   base output directory for output files (default: ['<Project Folder>\\output'])
  --log_type log_type   Input type of templates. (default: ['original'])
  --rules rules         Predefined rules to extract triples from templates. (default: None)
  --evaluation evaluation [evaluation ...]
                        Triples extraction evaluation metrics. (default: [])
  --openie openie       OpenIE approach to be used for triple extraction. (default: ['stanford'])
  --id id               Experiment id. Automatically generated if not specified. (default: None)
  --tag                 Tag variables in the output triples (i.e. [([variable])] ). (default: False)
  --save_output         Save the output of logs or templates triples. (default: False)
  --force               Force overwriting previous output with same id. (default: False

Examples

Only generate output from LogIE

This command only generates output from LogIE without evaluation using the provided online templates in their corresponding field of the json format specified above. The ground truth will be disregarded to generate this output.

python -m LogIE.run --templates "<Templates File Path>" --"<Raw Logs File Path>" --rules new --openie <OpenIE approach> --save_output --tag

Please note that --tag will tag the variables in the output triples (i.e. [([variable])] ).

Rank Summaries

python summarization.py --type Proxifier --model model/Proxifier.model --evaluate 1 --topk 5 #evaluation mode

WeibinMeng / LogSummary

readme