Prediction ahead time example

Hi everyone,

I have based on gym energy consumption example and try to make a future prediction.
I want to get 10 steps ahead from the last time value.
For example if my data has 1000 rows, and it steps has 1 minutes, i want to get 1010 which means:
1001 -> 1 minutes ahead
1002 -> 2 minutes ahead


First I have changed model params like below:
( It was mentioned in another topic. )

‘clParams’: { ‘alpha’: 0.019888046680498178,
‘regionName’: ‘SDRClassifierRegion’,
‘steps’: ‘1,2,3,4,5,6,7,8,9,10’,
‘verbosity’: 0},

But i did not get prediction, it is the same with result.inferences[“multiStepBestPredictions”][1] and result.inferences[“multiStepBestPredictions”][10]

I also could not manage to write csv file, all predicitons:

output.write([timestamp] , [consumption] , [p1] , [p2] , [p3])

Could you help me about this?

  1. How can I make future prediciton for example 10 steps from the last row
  2. How can I write to csv files all predicitons?

If you are getting data at 1000hz like you said, and you want to predict a minute into the future, you would need steps: [1000*60], which is not going to perform well. The TM only activates cells for the next predicted time step, it would be the SDR Classifier that tries to predict the probability of that value 60,000 steps into the future.

I don’t think you’re going to success with such small time slices and trying to predict so far into the future. Are you sure that sub-second patterns in this data really can be used to predict the longer scale patterns that you expect to see over a minute or two timeframe? Have you plotted this data out at this granularity to see if a human could identify patterns?

I dont want to get data 1000Hz actually.
MY problem only predict future example.
I can change my test data period.

by 1000 I only mean 1000 minutes total data.
And my step interval is 1minute.
So I want to guess 1000 + 1minute ahead,… 1000 + 10minute ahead.

And write them into csv file.

My advice is to plot your data at the aggregation you plan to analyze it. Show us the plot so we can see if this is a good case for HTM to provide value.