2019-03-18 09:53 PM
2019-03-19 01:26 AM
Hi @Community member ,
Custom data validation can be done on desktop and on target. As described in the X-CUBE-AI User Manual, in section 14.2 Custom data set file format?:
"The expected file is a simple text file in csv format with one flattened input tensor by line. The comma is used as a separator."
When you say the "process is not ending at all", is there any additional information displayed in the STM32CubeMX Output log window? (Window -> Outputs)
Best regards,
Guillaume
2019-03-19 09:31 PM
2019-03-19 09:32 PM
2019-03-20 02:17 AM
@Community member ,
Looking at your actitracker_raw.txt, it seems that your values are encoded as followed:
# Expected Class Nb, Expected Class Label, Timestamp, In X, In Y, In Z
33,Jogging,49105962326000,-0.6946377,12.680544,0.50395286
33,Jogging,49106062271000,5.012288,11.264028,0.95342433
33,Jogging,49106112167000,4.903325,10.882658,-0.08172209
Your custom dataset file should online include input tensors, that is, assuming your input tensor dimension is 3:
# In X, In Y, In Z
-0.6946377,12.680544,0.50395286
5.012288,11.264028,0.95342433
49106112167000,4.903325,10.882658,-0.08172209
X-CUBE-AI version 3.4.0 allows you to dump NN inference input and output data. If you look at your log when executing a validation, you should be able to see the creation of 3 files:
Creating C:\Users\<username>\.stm32cubemx\ai_valid_inputs.csv
Creating C:\Users\<username>\.stm32cubemx\ai_valid_outputs_model.csv
Creating C:\Users\<username>\.stm32cubemx\ai_valid_outputs_c_model.csv
Regards,
Guillaume