# constructing Feature Vector from given values

I am fairly new to signal analysis, so pardon any noobish questions, but I couldn't find a clear answer by googling.

I am using PyLab to calculate certain values from a given data.

My data is a 3D matrix of some sensor readings. For example, like this.

0.076614454;7.5848308;5.9759274

0.076614454;7.5082164;5.899313

0.30645782;7.201759;6.129156

0.30645782;7.1251445;6.0525417

I have calculated rms, mean, fft, etc. for them using various functions.

Now, I read somewhere to classify them, I would first require to make a "feature vector".

I didn't fully grasp the concept of feature vector, even though it's given in one of the comments to one of the other questions here on the site.

The question is, how do I make the feature vector given the individual values I got for this data.

How would my feature vector look like?

Can someone please demonstrate using a dummy example?

Any help is much appreciated!

-
Would this be more suited to Signal Processing SE? – Zev Chonoles Mar 30 '12 at 22:23
So you have a sequence of sense readings $\{r_1, r_2, \ldots\}.$ Each reading $r_i$ is a triplet $(v_1, v_2, v_3)$ of three values. This triplet $(v_1, v_2, v_3)$ is the raw feature vector for each reading. It's a vector since it's 3D. It's a feature vector since it describes the features of the reading $r_i.$ It's raw because it just contains the readings without any further processing. Constructing a sophisticated feature vector out of $r_i = (v_1, v_2, v_3)$ is something you need to read about in a textbook. – user2468 Mar 30 '12 at 23:08
Okay, I was trying it again and re-read your post. I got a bit confused. Earlier, I was taking the column vector for each of the x,y and z axis differently and calculating their mean, rms etc etc. but your posts says, it would be a triplet. A tuple of three values? How does that work? For example, how would RMS be calculated for a "triplet" together? – user27995 Mar 31 '12 at 12:35
@J.D.any help please? – user27995 Apr 1 '12 at 10:28

## 1 Answer

Choose your features based on what you want to analyze. If you want to analyze the sensor readings themselves you could use your raw data. In your case you would have four three-dimensional feature vectors. If you want to say analyze their frequency content you could use the wavelet coefficients, if it makes sense for your application. Generally speaking you use an appropriate transform and use the result as a feature. Then you perform your machine learning task on these vectors

-