Examples of Classifiers

Engineering Machine Learning
1

Hello,

I am searching for papers about the different classifiers.
I can only find YouTube videos… can anyone help me with that please? =)

Thanks a lot in advance.
Best,
Helena

2

If you meant classifiers used by HTM. That’s a good question…

The SDRClassifer in HTM.core/NuPIC is softmax regression

The old CLAClassifer is based on the properties of SDR. You can calculate the property based on this paper.

1 Like
3

Hey =) thanks for the fast reply.
Yeah it would be nice to have a paper about the CLA Classifier.
And whats about the KNN?

thanks for that… I’ll check that one =)

4

Sorry no… The only source I have is some old youtube videos and some discussion I have with the HTM.core devs. But there’s a good reason we don’t use it now. It is slow and not as accurate comparing to other algorithms.

KNN is just KNN, K-Nearest neighbor

1 Like
5

ok cool… Thanks a lot =)

6

I have another question.
I found that video:

In the video there are some figures, that seem to be part of a paper.
Where can i find that one? There is nothing mentioned in the slides.

And I am a little bit confused. Do we use the whole state of the HTM System for that task? I mean do we use the predicted cells as well as the active one? Or just one of these?

Thanks a lot in advace and a happy new year to everybody =)

1 Like
7

Hi @helena_Thielen,

Unfortunately, we only seem to have the video and code, but no slides or documents. :frowning:

From the author: “You can try SDRClassifier with predictedCells, but the default is set to use activeCells for several reasons. First, predictedCells may not even exist if TM hasn’t learned the sequence properly. And there may be a very large number of predicted cells if there are multiple predictions, so the sparsity of the predictedCells will vary more than the activeCells, which can lead to bad performance. Second, the information in the predictedCells should also exist in the activeCells, so SDRClassifier should be able to establish that mapping automatically.”

Happy New Years, and to everyone!

1 Like
8

hello =)
Thanks a lot for your help! :blush:

1 Like
9

Sooo hopefully my last Question on classifiers :sweat_smile:

When I am doing scalar prediction I know that I divide the whole range in Buckets.
And as I understand, these buckets can be seen as classes for the final classification/prediction.

But now I’m wondering: Say we have some buckets. One includes the values from 4-7.
Now the SDR-classifier returns that this bucket has the highest probability
So which value will be returned? I mean which value will be added to my csv file (or where ever I am saving the predictions)?

And will it always be the same value or is there any mechanism, that this can change over time?

Thanks a lot in advance.
Best, Helena

10

It is based on the moving average equation.

11

so this means the moving average of all scalars within a bucket?