The nail that sticks out gets hammered back in
- #1,009
- Jul 15, 2010 10:11am Jul 15, 2010 10:11am
- Joined Mar 2008 | Status: Still testing and trading | 1,454 Posts
- #1,011
- Jul 16, 2010 8:53am Jul 16, 2010 8:53am
- Joined Mar 2008 | Status: Still testing and trading | 1,454 Posts
- #1,013
- Jul 17, 2010 2:38am Jul 17, 2010 2:38am
- Joined Mar 2008 | Status: Still testing and trading | 1,454 Posts
Time for a little status update.
I've finished implementing support for all timeframes except weekly and monthly in my backtester. It takes 1.5 sec to load a currency pair (most of that being LZMA decompression). Switching between timeframes after the initial loading is almost instantaneous. The TFs are built from M1 candle files which are built from UTC ticks. I store what happened first (the low or the high), and you can see that displayed in the little points at the end of one of the candle wicks.
I also implemented basic workspace support, which can have major changes between them. For example, each workspace can have it's custom plugins. In the picture you can see a plugin which changes the color of the candle background to indicate the current detected trend.
http://dl.dropbox.com/u/190212/ff/workspace.png
One of the plugins can convert the M1 data to a different timezone. BTW, for performance reasons this is implemented in C#, which is dynamically compiled on the first use 
.NET is really nice here - 20 lines of IronPython to compile a .NET assembly starting from .CS source files and instantiate one of it's objects.
Animation with daily candles for different timezones:
http://dl.dropbox.com/u/190212/ff/timezone.gif
I've finished implementing support for all timeframes except weekly and monthly in my backtester. It takes 1.5 sec to load a currency pair (most of that being LZMA decompression). Switching between timeframes after the initial loading is almost instantaneous. The TFs are built from M1 candle files which are built from UTC ticks. I store what happened first (the low or the high), and you can see that displayed in the little points at the end of one of the candle wicks.
I also implemented basic workspace support, which can have major changes between them. For example, each workspace can have it's custom plugins. In the picture you can see a plugin which changes the color of the candle background to indicate the current detected trend.
http://dl.dropbox.com/u/190212/ff/workspace.png
Inserted Code
style:
chart:
draw_time_gap: false
candle_background_decorator:
type: Python
module: xxxxx
class: CandleBackgroundDecorator
bullish_brush: CCFFCC
neutral_brush: WhiteSmoke
bearish_brush: FFCCCC
error_brush: Magenta .NET is really nice here - 20 lines of IronPython to compile a .NET assembly starting from .CS source files and instantiate one of it's objects.Animation with daily candles for different timezones:
http://dl.dropbox.com/u/190212/ff/timezone.gif
Inserted Code
filter:
timezone:
type: C#
sources: base.sources
class: CandleTimeZoneConvertor
name: London Inserted Code
_TIMEZONE_MAP = {
"UTC": "UTC",
"Chicago": "Central Standard Time",
"New York": "Eastern Standard Time",
"London": "GMT Standard Time",
"Frankfurt": "W. Europe Standard Time",
"Bucharest": "GTB Standard Time",
"Tokyo": "Tokyo Standard Time",
"Sydney": "AUS Eastern Standard Time",
"New Zealand": "New Zealand Standard Time",
}
I've started implemented the modular aspect of the backtesting engine. When thinking about the design I quickly realized that it's much harder to do it with candles than with ticks, since for each node you would have to specify what value of the candle to use (like in MT - EMA on low, or EMA on close, ...). So I've finally bit the bullet and decided to drop candles and move to a time-series based engine. I also realized how I could easily replicate most candle-based trading strategy in it.
The actual streaming design is heavily based on Windows Media Foundation, since that is the third iteration of modular graph design Microsoft implemented (after DirectShow and DMOs) and I don't think I can come up with something better. More specifically I took the whole Topology layer and simplified it a bit since my needs are simpler.
http://msdn.microsoft.com/en-us/library/aa369306.aspx
http://msdn.microsoft.com/en-us/library/bb250382.aspx
I have already implemented a few nodes for testing:
And a simple manual chaining test to validate the design:
Now I'm looking into how to specify the actual topology and I've discovered the nice DOT language. The hypothetical graph spec I've posted a few posts back is actually a valid DOT language descriptor ) The ability to see the graph will be very useful to validate that stuff is properly connected together:
http://dl.dropbox.com/u/190212/ff/dot_simple.png
With a few annotations that can be made to look like this:
http://dl.dropbox.com/u/190212/ff/dot_decorated.png
Of course, the graphs are no longer fully valid since I've dropped the candles.
The actual streaming design is heavily based on Windows Media Foundation, since that is the third iteration of modular graph design Microsoft implemented (after DirectShow and DMOs) and I don't think I can come up with something better. More specifically I took the whole Topology layer and simplified it a bit since my needs are simpler.
http://msdn.microsoft.com/en-us/library/aa369306.aspx
http://msdn.microsoft.com/en-us/library/bb250382.aspx
I have already implemented a few nodes for testing:
PHP Code
using System; namespace Veloce.Sharp { public class NodeTransformTimeZone : Node { private Tick tick; private TimeZoneInfo timeZoneInfo; public override void init() { base.init(); inputReady = true; } protected override void updateConfig() { string id = cfg.getConfig<string>("timezone_id"); timeZoneInfo = TimeZoneInfo.FindSystemTimeZoneById(id); } protected override void takeInput(int inputIndex, Tick tick_) { tick = tick_; tick.time = TimeZoneInfo.ConvertTimeFromUtc(tick.time, timeZoneInfo); inputReady = false; outputReady = true; } protected override Tick generateOutput(int outputIndex) { inputReady = true; outputReady = false; return tick; } } }
Inserted Code
input_node = NodeInputTicks()
input_node.setTicks(ticks)
time_zone_node = NodeTransformTimeZone()
time_zone_node.setConfig(dict(timezone_id="GTB Standard Time"))
output_node = NodeOutputBuffer()
output_node.setConfig(dict(start_capacity=100000))
while input_node.canProcessOutput():
tick = input_node.processOutput(0)
print "> %s %f %f" % (tick.time, tick.bid, tick.ask)
time_zone_node.processInput(0, tick)
tick = time_zone_node.processOutput(0)
print " %s %f %f" % (tick.time, tick.bid, tick.ask)
output_node.processInput(0, tick)
ticks = output_node.getContent()
print len(ticks), "ticks"
for tick in ticks:
print " %s %f %f" % (tick.time, tick.bid, tick.ask) ) The ability to see the graph will be very useful to validate that stuff is properly connected together:http://dl.dropbox.com/u/190212/ff/dot_simple.png
With a few annotations that can be made to look like this:
http://dl.dropbox.com/u/190212/ff/dot_decorated.png
Of course, the graphs are no longer fully valid since I've dropped the candles.
- #1,020
- Jul 24, 2010 2:16pm Jul 24, 2010 2:16pm
- Joined Feb 2006 | Status: Blah blah blah | 1,410 Posts
The breaking of a wave cannot explain the whole sea.