I am using JFreeChart to do some backtesting in the Forex market. I have a csv sheet containing 1 years worth of 1 minute data. In all about 370,825 entries. I have loaded this data into a time series chart(obviously using Minute() for the time period), but the performance is horrible. I profiled my application and here are my results that use up all my memory:
org.jfree.date.SpreadsheetDate ->1,250,144 B
org.jfree.data.time.TimeSeriesDataItem ->625,840 B
org.jfree.data.time.Hour ->624,992 B
org.jfree.data.time.Day ->623,760 B
org.jfree.data.time.Minute ->623,648 B
This is just one chart. I need two more, but there is no way this will be possible. Is there anyway to reuse the objects needed for the data? As the subject title asked... Is JFreeChart a practical answer for this kind of application?
Thanks,
Scott
Is JFreeChart a practical answer
On a one-minute interval with 10 hours of data per day, 250 days/year that's about 150,000 points on the x-axis. My screen has 1280 pixels across the way so that would mean I would cram about 120 points into each pixel.
Put it another way, you only need to plot points every two hours over the course of a year to completely use the full resolution of the screen - in practice, for a year chart, each day's closing price would be perfectly adequate.
So basically, I'm saying that obviously the thing doesn't work because you are grossly overdoing the resolution.
Then again, maybe I'm mistaken - how big, exactly, is your screen?
Put it another way, you only need to plot points every two hours over the course of a year to completely use the full resolution of the screen - in practice, for a year chart, each day's closing price would be perfectly adequate.
So basically, I'm saying that obviously the thing doesn't work because you are grossly overdoing the resolution.
Then again, maybe I'm mistaken - how big, exactly, is your screen?
Actually it does work. Regardless of resolution, my performance is bad because of how much data objects I am creating (as seen from my previous post).So basically, I'm saying that obviously the thing doesn't work because you are grossly overdoing the resolution.
But I did do an experiment to test your little theory. I did a one month chart on my computer. I have two 20 inch monitors (1680x1050). I profiled each of the following: The first part I made my chart screen about half the size of one monitor and the second part I utilized both screens (3360) and the results.... there was absolutely nothing different. The performance and memory usage was exactly the same.
Thanks you for your response, but that's really not the answer to my question.
-
david.gilbert
- JFreeChart Project Leader
- Posts: 11734
- Joined: Fri Mar 14, 2003 10:29 am
- antibot: No, of course not.
- Contact:
The TimeSeries/TimeSeriesCollection classes are very flexible, but a little heavy on memory usage. In JFreeChart 1.0.2, a more lightweight dataset (org.jfree.data.xy.DefaultXYDataset) which uses a double[][] array as the underlying storage, was added. Maybe that will help you here. For the x-values, just use the "milliseconds since 1-Jan-1970" encoding defined by java.util.Date.
David Gilbert
JFreeChart Project Leader
Read my blog
Support JFree via the Github sponsorship program
JFreeChart Project Leader
Read my blog Support JFree via the Github sponsorship programI don't think you understand what I'm saying...
Your problem is that your data objects are too large and take too long to create - correct?
My point is that you can't see all the information in your charts anyway so you might as well make them smaller, then they'll be quicker to generate. A one-year chart with data points every minute has over a hundred data points in every pixel - so that's pointless resolution. In other words, you could reduce the number of points on the x-axis by a factor of 100 and teh chart would look *exactly* the same on the screen, but it would consume only 1% of the memory and fill 100 times faster...
Your problem is that your data objects are too large and take too long to create - correct?
My point is that you can't see all the information in your charts anyway so you might as well make them smaller, then they'll be quicker to generate. A one-year chart with data points every minute has over a hundred data points in every pixel - so that's pointless resolution. In other words, you could reduce the number of points on the x-axis by a factor of 100 and teh chart would look *exactly* the same on the screen, but it would consume only 1% of the memory and fill 100 times faster...
Size of X-axis in TimeSeriesCollection
how could I reduce the number of points on the x-axis by a factor of 100..in TimeSeriesCollection..
Kind request for reply
Kind request for reply
by retrieving less data and thus putting less data into the collection...
In our system I can simply tell the engine to give me points at specified intervals (minute, hour, day, for example) and I get a filtered set as a result.
If your system doesn't have something like that, maybe it's time to add it
In our system I can simply tell the engine to give me points at specified intervals (minute, hour, day, for example) and I get a filtered set as a result.
If your system doesn't have something like that, maybe it's time to add it