Heidi is upgraded to version 0.5

Written by Forex Automaton   
Saturday, 19 February 2011 14:40

Heidi has been upgraded to version 0.5. This upgrade focused on the memory consumption and is expected to stabilize the memory situation on the host.

In the ForexAutomaton systems, the accumulated "knowledge" of a system is represented by a set of objects describing time series structure at various time scales. To ensure adaptability, the objects have a limited life span going through the learning stage to the stage of active engagement in the forecasting process, before being eventually phased out, at which stage a new object is introduced into the queue. The life span of such an object grows with the time scale it describes.

In the versions prior to v0.5, a time would come regularly for several, and ultimately, all of the object queues, to be upgraded with new objects during the same time step. The result was that the system would go through a time step where it has to deal with a lot of memory demand, after the new objects had been created but the old ones had not been removed yet. This became a sensitive issue only on the rented host where memory comes at a cost, and remember, this system is a free service.

The solution was to stagger object queues with a scale-dependent delay so that there is never a point in time when all time scales undergo a simultaneous addition of memory. Version 0.5 is the first version where such a solution is implemented. This does not address the forecasting quality, only the resource demands of the system.

Meanwhile, work is underway to improve the forecasting quality for hourly close. The current approach is to take advantage of the daily seasonality effects by specializing the adaptive learning objects within specific time windows, and the early results look encouraging. Stay tuned!

Bookmark with:

Deli.cio.us    Digg    reddit    Facebook    StumbleUpon    Newsvine
Last Updated ( Saturday, 19 February 2011 15:04 )
 

World Time

New York: 22:42 London: 3:42 Berlin: 4:42 Tokyo: 11:42