This is the code repository of ConceptExplorer in Vast 2020.
Online system please visit http://101.132.126.253/.
We can only provide two real-world datasets due to certain confidentiality agreements.
Beijing Air Quality Forecast Datasets
paper: S. Zhang, B. Guo, A. Dong, J. He, Z. Xu, and S. X. Chen. Cautionary tales on air-quality improvement in Beijing. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 473(2205):20170457, 2017
datasets: https://archive.ics.uci.edu/ml/datasets/Beijing+Multi-Site+Air-Quality+Data
Movie Rating Prediction Datasets
Movie description: https://www.kaggle.com/rounakbanik/the-movies-dataset
Twitter dataset: https://github.com/sidooms/MovieTweetings
MovieLens 20M dataset: https://www.kaggle.com/grouplens/movielens-20m-dataset
Rotten Tomatoes dataset: https://www.kaggle.com/stefanoleone992/rotten-tomatoes-movies-and-critics-datasets.
After download datasets, you should preprocess the datasets and do some necessary operations such as Drift Level Index Calculate and The Consistency Judgment Model to get result that showed in the system.
Please refer to the paper for the specific calculation process. The final result should have the following structure.
dataset: {
'name': string, // name of dataset，eg: 'netflex'
'attribute': { // matrix of attribution
'negative': array,
'positive': array, //darray,
//shape = (batchNum, splitAttrSize, splitAttrSize)
}
'delay': { // results after get label
'time': [], // list, timestamp of each batch, saved by UNIX Time
'accuracy': [], // list, model accuracy of this batch
'nb_prob': [] // list, Bayes Predicted probability of this batch's drift level
'batch_delay': number // offset of nb_prob
'hit': [], // hit or not of this bacht(1 means hit, 0 mmeans miss)
'warning': [{
'start': number // start timestamp of this warning
'end': number, // end timestamp of this warning
'max_accuracy': [], // 1-Pmin of each warning
'max_accuracy_time': [], // timestamp of each 1-Pmin
'backend_accuracy '：[] // backend accuracy of each warning
}]，
'drift': [{
'start': number, // start timestamp of each drift
'end': number, // end timestamp of each drift
'max_accuracy': [], // 1-Pmin of each drift
'max_accuracy_time': [], // timestamp of each 1-Pmin
'backend_accuracy '：[] // backend accuracy of each drift
}],
'warningLevel': [ // threshold of warning level
'avg': number,
'max': number
],
'attributes': [{
'name': string, // Attribution Name
'predict': [], // Correlation between attribution and label of this batch
'correlation': [] ， // Correlation between previous batch and this batch
num: number // Dimension of this attribution
},
...// May have K-Attributions
]
}，
'online': { // Online result without label
'time': [], //list, timestamp of each batch, saved by UNIX ime
'weight': ndarray(num, attribute_num), // model weight, saved by numpy
'dataNum': [] // data num of each bacth, eg: The number of records of Netflix data in one day
}
}
// online_num = dataset.online.time.length
// delay_num = dataset.delay.time.length
// online_num - delay_num = C （const number) when delay_num > 0
The system consists of the frontend and the backend.
The backend provide data used in frontend.
This code can be used in Windows/Linux/OS environment with python3.
https://wiki.python.org/moin/BeginnersGuide/Download
Install required package
django >= 2.1
numpy >= 1.16
sklearn
Change directory to backend\, Run Command:
python manage.py runserver 8000
Backend server will start at your localhost:8000. You can change server address by command python manage.py runserver [ip]:[port]
This code can be used in Windows/OS environment.
- node.js
- yarn
Please add node.js into your environment variables
Change directory to frontend\,
First install required package:
yarn install
Then, start:
yarn start