Time series anomaly detection with Chronos2

Welcome to the Chronos2 time series anomaly detection demo! This application allows you to upload your own time series data and visualize the detected anomalies using the Chronos2 pipeline.

Instructions

1. Click on the Upload Time Series Data button to upload your time series data in CSV format. The CSV file should have as columns only:
   • "timestamp": the timestamp column of your data (e.g., "2023-01-01 00:00:00"). It is optional.
   • "values": the columns containing the values of the time series data. They can be named as you wish. At least one column of values is required.
2. Answer the question about the timestamp in your data to help the model understand the temporal structure of your data.
   • if present, you will need to specify the column name of the timestamp in your data.
   • Otherwise, no need to do anything, just mark No.
3. Click on the Detect Anomalies button to run the Chronos2 pipeline and visualize the detected anomalies.
4. If the number of series is reasonably small, we will plot the original time series along with the detected anomalies.
5. We will provide a downloadable CSV file containing the original time series data along with an additional column indicating whether each point is an anomaly or not. We will label as 1 anomalies, as 0 normal points and as -1 the points for which we don't have a prediction because they are before the minimum length required by the model.

Note

• The Chronos2 pipeline is designed to handle multivariate time series data, so you can upload datasets with multiple columns of values.
• For a correct prediction, a minimum length of 64 data points is required. If your time series is shorter than this, the pipeline will not be executed.