[5.4 bug, fixed]
Author: sf631
Creation Date: 9/21/2009 12:28 PM
profile picture

sf631

#1
I am running backtests of a strategy in WLP 5.4 and getting two error messages that I've never seen before in this context. The strategy is using ASCII datasets I've created which have closing price plus some additional data fields needed to run the strategy.

The first is an error message that causes WLP to crash on the final symbol of the dataset. It reads:

CODE:
Please log in to see this code.


Doesn't matter what the final symbol is, it will always choke on the last one and shut down. I've tried this with the full dataset (1500 symbols), reduced dataset (400 symbols), and a very small dataset (30 symbols). Crashes happen on all but the smallest (30 symbol) datasets, which run fine.

The second error I'm getting is when trying to set a data range for the strategy. If I use anything but "All Data" I get a runtime error saying that I can't mutatea value collection derived from a dictionary. The strategy runs fine on "All data" but I don't understand why I can't just test a few weeks of my time horizon. If I pick a timeframe longer than the data available it works fine, just as if I chose "all data". BTW, there are about 60 daily bars.

Here is the gist of the code:

CODE:
Please log in to see this code.


Any ideas? Thanks
profile picture

sf631

#2
Also, somewhat related, is it possible to use the close price for a symbol from a regular (fidelity/yahoo etc...) dataset while using the extra fields from the ASCII instance of the same dataset? Basically I'd like to run the test on the dataset made up of the ASCII dataset, but use price info from another place so I can be sure the data isn't biased



profile picture

Eugene

#3
QUOTE:
If I use anything but "All Data" I get a runtime error saying that I can't mutatea value collection derived from a dictionary.

Sorry for the inconvenience Chad. This is a Wealth-Lab client version 5.4 bug affecting the ASCII provider, if the data happens to have custom fields defined e.g. Open Interest. It's fixed in 5.5.

I'll take a look at the 1st issue.
profile picture

sf631

#4
Thanks Eugene. BTW, I just added symbols to the smallest dataset and verified that the problem exists at 70+ symbols as well. Looks like somewhere in the mid double digits that is the threshold point.
profile picture

Eugene

#5
I haven't experienced the first issue you describe so far in this connection. To reproduce it now, took a dataset of 500 ASCII symbols with custom fields defined, modified your strategy so that it would point to the right named series, limited the number of bars loaded, and it didn't crash for me at all and no error messages were generated (I deliberately used 5.4 instead of 5.5).

Looks like you need to closely inspect your data for errors. I could take a look if you share the ASCII data for one of such failing symbols somewhere (e.g. by creating a new support ticket and attaching). No need in whole dataset, I'll just clone the file to test.

QUOTE:
Basically I'd like to run the test on the dataset made up of the ASCII dataset, but use price info from another place so I can be sure the data isn't biased

I think it's possible if you use an overloaded call to GetExternalSymbol that accepts a DataSet name as one of its parameters. This way, Wealth-Lab returns a Bars object from which you'll need to extract the custom fields.
This website uses cookies to improve your experience. We'll assume you're ok with that, but you can opt-out if you wish (Read more).