OpenViBE forum

Hello everyone,

I have a question regarding the feature extraction in mi-csp-3-classifier-trainer.xml scenario.

Consider the following diagram:

..................................................................................40 trials
................................................................................. /.........\
.................20 (right) epochs and each epoch has a duration of 4s...........20 (left) epochs and each epoch has a duration of 4s.
......................................................................................|
......................................................................................v
................................................. split the 4sec signals to chunks of duration = 1 sec for each 1/16 of second.
................................................. since 1 sec has 16 chunks => 4 sec -> 64 chunks (for 1 epoch)
................................................. since 1 epoch has 64 chunks => 20 epochs have 1280 chunks
................................................. so each feature aggregator box constructs a feature vector with 1280 features
................................................. and therefore, classifier trainer box gets as input 2 feature vectors with each one having a size of 1280

Why when we execute this scenario, we get on the console that we have 1080 feature vector(s) for input 1 and 1080 feature vector(s) for input 2?

I would appreciate your help.

Thankyou,
Sophia

Hi Sophia,
Your chunk compute is false, you must have 1s entierely so it's not (16*4) 64 but (16*3+1) 49.
But It's strange because to have 1080 features vectors, it's with 54 chunks by trials (1080/20) You have 100 chunks too many.
You can Test to wirte a csv with Features of one side with Stimulations and check the time, maybe you can see a strange behaviour.
Thibaut