Getting started

To get started, fork the repository from github.

Environment structure

The repository contains the following folders and files:

• docs contains the files for generating documentation in Sphinx

• example contains example files for Phase 2 submission

• src contains the starter code

• src/utils contains utility functions

• src/leak_detection.py python class you should implement and submit

Challenge

Implement your solution in leak_detection.py under the method predict. The method takes in a list of features and should return a boolean. It should return true if leak is detected and false if there is no leak. You can create new files and import them into predict.py. Example of how predict method will be called is in evaluate.py. You can use it to test your solution.

Set up environment

To set up environment you just have to install all the dependencies. You can do so with pip:

pip install -r requirements.txt

Phase 1

The first task goal is to analyze and try to detect water discharges based on unlabeled data from micro-turbine signals.

You have to use advanced anomaly detection algorithms through machine learning or advanced statistical analysis to predict leakages.

Submission

You have to upload json file with periods labels, where the leakage occurred.

{
    "prediction_results": [
        {
            "file_name" : "my-file4.csv",
            "end_periods": [360, 720, 10800, 14400],
            "leakages": [1,0,1,0]
        },
        {
            "file_name" : "my-file2.csv",
            "end_periods": [1800,5400,8700, 14400],
            "leakages": [1,0,1,0]

        }
    ]
}

Phase 2

Before participating in phase 2, you should finish phase 1.

The second phase has more advance dataset, than the first one. The labeled data (train dataset) also have seasonality component, but the principle of leakage remains the same as it was analysed in phase 1. Additionally, the trend could be easily analysed on dataset samples, describing time series data signals, obtained from micro-turbine IoT device.

Beside the resulting json file, participants will also have to submit the working algorithm, which is able to predict real time stream data. Such prediction algorithm will be tested and evaluated in EvalAI platform.

After submitting final submission, the algorithm will be evaluated in live stream data, which will also present the final result, and will be presented as final challenge score.

Submission

The result.json file should look something like this:

{
    "prediction_results": [
        {
            "file_name" : "my-file4.csv",
            "end_periods": [360, 720, 10800, 14400],
            "leakages": [1,0,1,0]
        },
        {
            "file_name" : "my-file2.csv",
            "end_periods": [1800,5400,8700, 14400],
            "leakages": [1,0,1,0]

        }
    ]
}

When submitting the algorithm, you should also include the requirements.txt and all the necessary files for the algorithm to work. You can submit everything in a zip format. An example of the submission files can be seen in example folder (file tree.joblib is optional).

Submitting

To submit your solution you need to upload it to our EvalAI server.

Solution documentation

After completing the tasks, please add documentation of your solutions to solution.md.