moving_mahalanobis

Github repo for moving mahalanobis model

Data

The scripts read the data in JSON format. The data structure is as follows:

{
  "serie_name_1": {
    "X": [list containing the values of the series],
    "y": [list containing the labels of the series (1 for anomaly, 0 otherwise)]
  },
  "serie_name_2": ...
}

Instructions for Processing Data (Non-JSON Format)

To handle data that is not in JSON format, follow the steps below:

1. Download Data
   Download the data from the various datasets.

2. Organize Data
   For each dataset, place the data files in the corresponding folder.

3. Run the Data Loader Script
   Execute the following command in your terminal:

   python ./{dataset}_anomaly/{dataset}_loader.py
   

Models

To run the model, choose the parameters on the file config.yaml and then run:

python ./{dataset}_anomaly/{dataset}_loader.py

python main.py

To run the other models, refer to the scripts_{model} folders and follow the instructions in the README.md file.

Print results

When you run the script main.py, the results and the scores (optional) will be store on the folder results. Use:

    python print_results.py --path [path to json file of the results] --n_decimals [2]

To display the scores on the consol. You can eval the model with another window size by running:

    python eval_from_scores.py --path [path to json file of the scores] --seq_len [custom sequence length]

Model presentation

Please find a summary report of the project in summary_report.pdf and the full report (in French) in full_report (french).pdf