Effect of disableLearning

Hi, all.

I’m confused with the role of “disableLearning” function during OPF.
I have a 180 thousand records in a csv file that I want to run with in OPF.
I called model.disableLearning() at certain record, for example 1,000th record.

After calling this function, predicted value was fixed with a number.

2016-09-06 13:03:00,6.5,6.501031328199021
2016-09-06 13:04:00,6.7,6.560721929739314
2016-09-06 13:05:00,6.7,6.542505350817519
2016-09-06 13:06:00,6.7,6.589753745572263
2016-09-06 13:07:00,6.7,6.6228276219005835
2016-09-06 13:08:00,6.5,6.6459793353304075
2016-09-06 13:09:00,6.7,6.662185534731285
------------------------------------------------------------------- disableLearning called here
2016-09-06 13:10:00,6.7,6.613529874311899
2016-09-06 13:11:00,6.7,6.613529874311899
2016-09-06 13:12:00,6.7,6.613529874311899
2016-09-06 13:13:00,6.7,6.613529874311899
2016-09-06 13:14:00,6.7,6.613529874311899

2016-09-26 10:38:00,6.3,6.613529874311899
2016-09-26 10:39:00,6.1,6.613529874311899
2016-09-26 11:23:00,6.3,6.613529874311899
2016-09-26 11:24:00,6.3,6.613529874311899
2016-09-26 11:25:00,6.3,6.613529874311899
2016-09-26 11:26:00,6.3,6.613529874311899
2016-09-26 11:27:00,6.5,6.613529874311899

The value (6.61352987…) continues no matter the actual value is.

Thereby, I’m curious the role of disableLearning.
My understanding was it can do some prediction based on the learned pattern up to 1000th record.
Is it natural to see such a fixed prediction value after disabling learning?
Please help this and my other previous question, too… no reply more than 10 days… T.T

1 Like

@oreore did you disable learning after only 1000 records? That is probably not enough for the system to learn patterns. Also, what are your encoder parameters? Looks like your data is by the minute. That’s only 16 hours of data, so you won’t have learned any daily patterns in the data at all. It could be that the system has not seen enough data to learn the patterns by the time you disable learning.

Sorry for late catch up…!!
Thanks for your advice.

Actually, before I upload this problem I tried several different record positions for putting the disableLearning.
This phenomenon was observed even though I put disableLearning after presenting a longer records.

But, once I change the resolution of RandomDistributedScalar encoder from 0.001 to 0.1, such observation was gone away (predicted values are not fixed…) even calling disableLearning after only 1000 records.
I guess this observation was a result of particular settings of encoder parameters.

Anyway, it is good to hear from you that this observation is generally NOT expected result of disableLearning.
Thank you.