Stop learning and Make prediction

I am new to HTM and I started out with learning swarming technique followed by cortical learning algorithm . I am totally convinced with the working of swarm , implemented it and got the model which consist of the various components used and related parameters for each of them. Now I implement the CLA model( i.e, by running, . Now Correct me if I am wrong that At Tth time level it is predicting the value for (T+r)th time level and accordingly getting me the error but I want to stop learning at some point . Basically I want to run it over a dataset in which output is not given and it will itself predict the value of the output.

My goal is to use it for any type of problems (ie., regression , classification etc.) where learning is restricted to some point and then it will give the output by itself.

It will be very helpful if anybody can make it clear to me because I was using swarming with the goal of making predictions for which output is not given after some time.
Please try to explain it is simple words even if it is a stupid doubt

Thank you

1 Like

It all depends on how the model was set up. It could be predicting 1 or 100 steps into the future. If you are following one of the “hotgym” tutorials, it is probably predicting one step (15 minutes) into the future.

I’m not sure what you mean by that? If the mode is a TemporalAnomaly model (check the inferenceType in the model params), you will be provided an anomalyScore within the model results. This is not an error, but an indication of how comfortable the model is with the data.

I don’t understand what you mean by this. You want to train a model on data, then turn learning off (which is possible) and stop sending in the predicted field’s data? This isn’t how NuPIC works. In order to get a prediction, you have to send in a value. For each row of input sent to the model, a prediction is made. If you don’t send input, there’s no prediction.

Have you read our main wiki?

You have some basic questions that should be answered in this wiki. Can you take some time to read through it and watch some of the tutorial videos like the Hotgym tutorials?

@priyankyadav31Hi There,

We don’t yet have a concrete route from zero knowledge to working familiarity yet. Most users read Jeff Hawkins’ book “On Intelligence”, and that’s how they start to resonate with the theory and its practices and implementations. The book discusses the importance of “prediction” and how it relates to our theory of machine intelligence - its a discussion of what intelligence actually is and how it can be described as a hierarchical memory system. There really isn’t any substitute for reading the book (which by the way is not big; has bit letters and is very very accessible) :stuck_out_tongue:

From there, Raul’s video is my next recommendation - because the book doesn’t talk about SDRs and such. After that I recommend Matt Taylor’s (@rhyolight) new HTMSchool videos to dive in more deeply on some of the components (Note: Those videos are a Work in Progress, and haven’t been developed for all the major topics yet. They are being produced in order though, so watching them now is essential)

It is important to get a 20,000 ft (or meters depending on your culture origin) view of what is being done before diving into specifics?


Thank you Matt for your kind answer .
I think I am not able to convey my doubt clearly I was just asking if the output sequence is repeating over and over again thus after some time no need to learn anymore. So after that time I want it to work over same type of dataset but for which output is not given and want CLA to predict the exact output .

And ya…by the error in my previous post means altMAP error found by comparing the predicted and the exact value…

From now I will try to recover and improve my basics before asking on this forum

Thank you for your precious time

If the pattern is learned, and you turn off learning, then yes the model will still remember the pattern. This is useful if you want to train a model on a specific pattern, then compare it to new patterns to see how different they are from the learned pattern using anomaly scores.

My intention was not to reprimand you, only to point out resources. :slight_smile:

1 Like

Thankyou Matt…:grinning:
I got it right…It is really helpful…I have sought it out


Incidentally, my intention was only to help guide you to the best of my ability too - using what worked for me. :slight_smile: