Hawkular - Hawkular Data Mining 0.1.0.Final Released

I am happy to announce the first release 0.1.0.Final of the Hawkular Data Mining project. It contains several time series models and utility classes for time series modelling.

Time series models

Simple exponential smoothing
Double exponential smoothing (Holt’s linear trend)
Seasonal triple exponential smoothing (Holt Winters)
Simple moving average (Weighted moving average)
AutomaticForecaster - which automatically selects the best model
Parameters of all models are estimated using maximum likelihood estimation and models are designed for online learning

Time series manipulation & Statistics

Augmented Dickey-Fuller test
Autocorrelation function (ACF)
Time series decomposition
Time series lagging
Time series differencing
Automatic period identification

Integration into Hawkular

The integration into Hawkular can be found in datamining branch in the main Hawkular repository. Predictive charts are located in Explorer tab. Predictions can be enabled for any number of steps in the future for any metric being collected. Prediction engine automatically selects the best model for given time series. Currently it selects from simple, double and triple exponential smoothing models. In the following charts predictions produced by these models are showed.

Figure 1. Model Triple exponential smoothing

Figure 2. Model Double exponential smoothing

Figure 3. Model Simple exponential smoothing

Note that currently there is an issue with querying historical data from Metrics due to authentication issues. However Data Mining still receives metrics from bus so it is important to enable predictions as soon as possible and wait for some time to collect data (or increase collection interval). This will be fixed in the next versions.

Hawkular Data Mining 0.1.0.Final Released

Time series models

Time series manipulation & Statistics

Integration into Hawkular

Next Steps

Links

Thanks