Autolearn question.

If you run the correction function within wret for say the last 30 days a bias is calculated and applied to all future forecasts. If you were to run the correction feature again one week later again for the last 30 days would you not start running into issues. 23 days of forecasts will be made with no bias however the most recent 7 days would have forecasts with bias adjustments.

Having part of the data bias adjusted and some not would this not lead to incorrect bias adjustments. The more corrections attempted the worse it will get…

Clearly I am missing something here and my logic is flawed…

Sent from my GT-I9505 using Tapatalk

Good question, and - if I understand the question correctly - I think I have the answer for you:

The biases that were used with any forecast are recording IN the forecast data (.wxf files) and are taken into consideration in the analysis. In fact, you see this explicitly in details.txt, where it reports the average of the previous values of the bias corrections used in those forecasts. This means it makes little if any difference whether or not bias corrections were used, or even whether ro not they were particualrly good ones. They are factored into the analysis anyway.

As a (very oversimplified, considering there or four types of corrections being applied) example, suppose you ran an initial 30 day analysis on a bunch of forecasts made with no bias corrections, and the forecasts were, on average, 2 degrees too cold. Then, future forecasts (until another analysis is made) would effectively have 2 degrees added to them to correct this. Now suppose that the first 15 days had no bias correction applied, but the second 15 had a correction of 2 degrees (warmer) applied. wret would then know that, on average, a +1 degree correction was applied to the data. If, then, it found the forecasts 1 degree too warm, it would apply a 2 degree correction to raw output.

I hope this answers your question. Please elt me know whether it does, and whether I understood your question correctly.

Barring some yet uncovered bug, I think I have a robust analysis method going here. My remaining questions revolve around how best to use it. I am looking at my own data and that of other users. It appears the main problem is how to balance the competing interests of responsiveness to recent pattern changes (requiring a short analysis period) and the need for enough data to make the corrections statistically sound. I’m currently trying 45 days, with recent weighted more heavily.

Tom

Thanks Tom that answered my question.

Sent from my GT-I9505 using Tapatalk

I'm currently trying 45 days, with recent weighted more heavily.
I may be misunderstanding or simply missed something...but, how do you 'weight' more recent or older forecasts?

Hello to everybody!!

I

Settings in autolearn are saved on normal exit from the program, not as soon as they are made. So, once you have your settings re-entered, just exit autolearn and restart it. The settings should then survive a computer shut-down.

As for the question two posts above, about weighting recent forecasts more, that’s a check box option in wret.exe, and autolearn follows that instruction. I have not yet decided whether it’s best to use it or not. Right now, I am using it, but will uncheck it sometime soon and let that run for a while to see if there is any difference in accuracy.

Hope that helps!

Tom

I have not yet decided whether it's best to use it or not

Then I shall leave it alone, thanks Tom.

Hi Tom,

Sorry for my delay thanks cause I

Hello,

I have two questions.

First some times (the most) at the manually start of the
Autolearn take the message Run-time error

Is the autolearn error actually a message from wret.exe? If so, I recently solved a problem for another user, in which there was a problem with the WD log files. I made wret.exe able to handle this. Your problem might be the same thing. Get this version:

www.wxsim.com/wret.exe

and use it to replace ther one you have and see if it fixes the problem.

With the Soil Data problem, I think you my be trying to access the form to set things up, with auto run on. Go to Start/Auto Run, and Disable the scheduler and try again. Then after you have made settings (you may be OK just using the defaults if yuo don’t know soil type, etc.), check the ‘Allow soil calculations’ box at the bottom, and click OK to leave the form. Then, go back to the Auto Run form, and check the Soil box there before re-enabling the scheduler.

Let me know how that goes.

Tom

Thanks Tom,

The Autolearn problem solved making a reboot to my pc. The new Wret version I had installed before. Also I use Weather Link and not WD.
As concern the problem to Soil data and this solved as you telling me. From the Auto run I disable only the schedule without uncheck the soil data box. Then
I check the autorun again.

Everything is OK now.

Thank you again.