2026-03-15 6:41 AM
Hello,
I am using MEMS Studio (latest version) with the LSM6DSV320X to configure the Machine Learning Core with 2 decision trees.
Steps I follow:
Open MLC tool, set workspace via Browse
Import mlc_settings.json — data patterns load correctly
ARFF Generation tab — add features, click Generate ARFF File — generates features.arff successfully
Decision Tree Generation tab — set 2 trees, assign labels to each tree
Click Generate Decision Tree
Error:
Unable to open features_DT0.arff
I notice two problems happening:
MEMS Studio generates features.arff but then looks for features_DT0.arff when generating 2 trees. With 1 tree it works fine.
When the error occurs, the file features_DT0.arff (which I manually copied/created) gets deleted from the workspace folder.
After the error, mlc_settings.json gets reset — the dectrees[].filename fields are cleared back to empty strings, and sensor settings revert to defaults.
What works:
1 decision tree generates successfully every time
What fails:
2 decision trees — always fails with the above error
Questions:
When using multiple decision trees, should the ARFF be named features.arff or features_DT0.arff? Is the naming scheme documented?
Is there a known issue with mlc_settings.json being overwritten/reset during a failed generation?
Is there a way to call the underlying mlc_configuration_tool CLI directly with a settings file to bypass the GUI?
Device: LSM6DSV320X
OS: macOS
MEMS Studio version: latest available
2026-03-26 3:46 AM
Hi @Rickflor17 ,
We managed to reproduce the issues reported:
1) Unable to generate features_DT0.arff
The issue is platform dependent and occur on macos and we will fix it in next MEMS Studio release. When user requests to train a decision tree using a different set of labels compared to the ones used to generate features.arff, an additional arff file is generated in the workspace with file name feature_DT$${dectree_num}.arff and the new arff file is used to generate the dectree. This file generated by MEMS Studio and managed by the tool. We will improve the documentation to clarify this behavior.
(As workaround to proceed the development we suggest to test on Windows if possible)
2) After the error, mlc_settings.json gets reset — the dectrees[].filename fields are cleared back to empty strings, and sensor settings revert to defaults.
We were not able to replicate this issue. The only action which clears decision trees set is "Generate ARFF" request which forces the decision trees list to be restored to empty assuming that user wants to train decision trees using the new .arff file generated. We will proceed with our analysis and try to replicate the issue reported or review the flow.
Since there is no way to reproduce the same flow without the file processing performed on MEMS Studio, if you cannot use a Windows machine, we would suggest you to generate a single decision tree using only the files with the labels you need (single decision tree flow repeated multiple times) and as a last step generate the configuration importing all the decision trees generated and mapping features according to the features configured in ARFF generation page.