Anomalie detection on Shapelets

Question

I have been using Shapelets recently for my work (mostly the dataapp) and I was wondering how we could use the matrix profile pattern recognition in the dataap for my time series?

If anyone can help me on this, that would be appreciated!

Answer 1

For using anomaly detection on the data you need to use the stumpy library and then create a function that will be on top of you time series. Here's an example of a stumpy function:

import pandas as pd
from shapelets.apps import DataApp
from shapelets.apps.widgets import LineChart, View
import stumpy
import numpy as np
import datetime

def Anomalies(ts: pd.Series, windowSize: int, top_k: int) -> LineChart:
    # Find discords
    mp = stumpy.stump(ts, windowSize)

    discords_idx = np.flip(np.argsort(mp[:, 0]))

    views = []
    for m in discords_idx[0:top_k]:
        views.append(View(start=ts.index[m], end=ts.index[m]+datetime.timedelta(milliseconds=windowSize)))
    return LineChart(title='Anomalies', data=ts, views=views)

This function takes a Pandas time series ts, a window size windowSize, and a number of top anomalies top_k as input. It then uses the stumpy.stump() function to compute the matrix profile for the time series. The function then identifies the indices of the top k discords in the matrix profile and creates a list of View objects representing the start and end times of each anomaly. Finally, the function returns a LineChart object containing the original time series data and visualizations of the identified anomalies.

You can call this function in your data app to quickly identify and visualize anomalies patterns in your time series data.

you can look at your documentation https://shapelets.io/doc/ https://github.com/shapelets/shapelets-demo