Configuring the Meta classifier whith UNICO

[DGiov.1]

Hello,

I'm working on an ultra low power application where the MLC of the LSM6DSO32X is used to wake up the mcu when interesting stuff is happening, in particular I implemented two decision trees, each one focused on the detection of a pattern.

In order to reduce the false wakeups, I would like to apply the meta classifier only on one of the two trees, for the other one the counters are set to 0.

Now the problem is that I don't understand how to do it with Unico. At some point of the MLC configuration in Unico, it ask for the end counter for the meta classifier (see attached image), however there is no way of configuring the sub-group of the meta classifier (I would like the counter to be 1 for the no_change result and 10 for the change result). It looks like Unico groups the DT result base on some internal rule that I cannot control.

This is the meta classifier configuration in unico, I can only set the counter but I cannot configure the sub-groups.

 

I would like to specify that the decision tree of this project are 'manually' designed and not 'trained'. I took the Motion intensity example in AN5656 as a guide and my decision trees are very very similar to that one.

Are maybe the sub-groups configured in the decision tree file?

[Eleon BORLINI · Accepted solution]

Hi @Davide G_Oiovanelli​ ,

please consider this part below that we added to the Github example for multiple decision trees for your and similar purposes:

When two or more decision trees are configured, Unico asks to specify the list of labels for each tree, separated by semi-column. The corresponding numeric output is the position index (starting from 0). Dummy labels can be introduced in order to assign a different numeric output to a given label, as described in the following example:

• Decision Tree #1 with numeric output described above (values 0, 1, 2)
  • Input for Unico: stationary; medium_intensity; high_intensity
• Decision Tree #1 with different numeric output (values 0, 4, 8)
  • Input for Unico: stationary; unused; unused; unused; medium_intensity; unused; unused; unused; high_intensity
  • This second case (values 0, 4, 8) allows separating classes in different meta-classifier sub-groups.

If my reply answered your question, please click on Select as Best at the bottom of this post. This will help other users with the same issue to find the answer faster.

-Eleon

[DGiov.1]

Thank you Eleon, I'll try it soon.