README.md

Requirements

Python 3.8.0

conda create --name UniParser python=3.8
conda activate UniParser
pip install -r requirements.txt

Step 0: Download Glove Embedding

Download from this link () and put it in the correct directory

wget https://apache-mxnet.s3.cn-north-1.amazonaws.com.cn/gluon/embeddings/glove/glove.6B.zip

Unzip the file and add a line in the first line of the embedding file as follows, to transform it from glove format into word2vec format

400000 50

Here 400000 is the number of lines of the embedding file, and 50 is the embedding dimension.

Run and evaluate UniParser on Loghub-2k

conda activate UniParser
./run_2k.sh

Run and evaluate UniParser on Loghub-2.0

Please notice you need to download the full_dataset of Loghub-2.0 first.

conda activate UniParser
./run_full.sh

Evaluation of LSTM-based log parsers (UniParser)

Since the performances of LSTM-based log parsers are significantly influenced by the delimiters, and the authors of UniParser do not provide the detailed delimiters. Therefore, for UniParser, we only use space as delimiter, and temporarily change the evaluation of Parsing Accuracy: (We check each token after tokenize, and any token found to contain <> is consequently deemed as <>). This is an imprecise estimation of PA, yet it ensures the general trend is accurately represented. Others employing UniParser may leverage other delimiters to get a more precise measurement of performance.