Unico & Multiple Decision Trees & UCF - process to use externally generated Decision Tree files

[BWats.1]

Hi,

Using UNICO (v9.10.0.0), the process to generate a UCF file with multiple Decision Trees seems to be:

1. Load all the test data for ALL Decision Trees, labelling each data set as you would for an individual Tree (presumably the labels need to be unique across all trees)

2. Use the MLC Configuration Tool

3. When it gets to the part about multiple decision trees, enter the number required.

4. Select all the Features required across all the Trees

5. For each tree, select the labels entered in step 1 that are to be grouped into that tree

6. At the generate step it will generate a Decision Tree file for each Tree, and then only one UCF file.

Is that correct?

What are the implications/constraints/issues when generating a UCF with multiple Decision Trees if one or all of them are generated external to UNICO, or if a UNICO DTree has been modified (eg a Node value/threshold changed)?

• Do all the Features specified in the Decision Tree files need to be the same and in the same order? (ie listed in the DTree file after "features:")
• Do the Feature instance numbers (ie "x" in Fx_<feature>_<axis>) have to be aligned (eg F7_PeakToPeak_on_ACC_V)

Thanks,

Brendon.

PS Had the multiple Decision Tree process changed from Unico V9.8?

[Eleon BORLINI]

Hi Brendon @Community member​ ,

You can find HERE an example about how to combine 2 different Decision Trees, described step by step from the log acquisition to the "final test".

Moreover, to answer to the second group of questions, in this example Matlab is used to show the possibility to use an external tool, compatible with Unico through dedicated scripts provided by ST.

No change expected in the MLC tool between Unico 9.0 and Unico 9.8.

Let me know if this example can help toy and, if so, 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

[BWats.1]

Hi Eleon,

Thanks, yes the example is useful.

However it seems the steps have changed in Unico v9.10.0.0 (I updated my post later with the Unico version I was using).

The ARFF files are now generated separately and the choice to generate a Decision Tree file or import an externally generated file is offered when the GENERATE button is presented. It seems, prior to Unico v9.10.0.0 (or earlier?) there was no option to generate the Decision Trees when multiple DTs were requested. It would be useful to have that example you linked to updated.

I gather it is important the imported Decision Tree files both have the same set of Features, numbered the same, even though they are not all used within the Tree.

Also, it would also be good to have confirmation on these, and any other details, on the data sets used:

• The ODR rate and full scale values need to be the same at the time of data sampling, and in the MLC configuration settings
• The sets of data collected to be used in one of the Trees, needs to be of a similar duration
• The MLC window should be chosen to cover the maximum duration of the actions (or action transitions) to be detected.
• The sets of data exclusive to a Tree don't need to be of the same duration as data for other Trees.

BTW - are there any more detailed release notes on Unico beyond that in the UM1049 manual? It seems UM1049 has not been updated since revision 6 in early 2019. It would be useful for UM1049 to link to some examples like the one you referenced.

Thanks,

Brendon.

[Eleon BORLINI · Accepted solution]

Hi Brendon @Community member​ ,

I double-checked internally and, well, you are right.

From Unico version v9.8.9.0 to v9.10.0.0 the management of multiple decision trees has been improved and simplified as follows:

• Besides the complete .arff file (with all the features), now alo another .arff for each of the decision tree is generated, in order to simplify the generation of the single decision trees.
• The generation of the single decision trees can be performed now either directly with Unico or with external tools (in that example I showed you, Matlab was used)

Our experts will evaluate the possibility to update the example on Github linking Unico new versions, or alternatively to create a new example.

Coming to your other questions:

• The ODR rate and full scale values need to be the same at the time of data sampling, and in the MLC configuration settings --> This coherence would be the best case.
• The sets of data collected to be used in one of the Trees, needs to be of a similar duration --> ya it should be better, but the important is to define the same time window.
• The MLC window should be chosen to cover the maximum duration of the actions (or action transitions) to be detected. --> Right, but also a "silent" label would be useful to catalogue the "idle" state.
• The sets of data exclusive to a Tree don't need to be of the same duration as data for other Trees. --> Right, also internally to the same decision tree the duration of the acquisition might slightly vary.

-Eleon