Validating custom dataset on STM32 X AI.

[AS.4]

Hi

I am following this material to start using STM32 for AI. I have got to the data validation step on page 29 of the document. It works fine while validating using random numbers.

But while trying to validate using the dataset, I get the error:

TOOL ERROR: cannot reshape array of size 1317858 into shape (219643,90,3,1)

Could I please know what should be the validation dataset?

[Romain LE DONGE]

​Hello,

What dataset are you using to validate your model ?

From what I can see, it seems that there is a shape problem in your dataset: (1317858/3 = 439286; 439286/90 = 4880.955 => not an integer), could it be possible that samples are missing maybe ?

Looking at the shape, it looks like an array of size 59303610 (219643*90*3*1) is required

Best regards,

Romain

[AS.4]

Hi Romain

I am using a dataset that I collected on-site from a mmWave RADAR.

I'll check the dataset to determine whether or not samples are missing.

Could you please recommend the train: test: validate data size ratio that should be used?

Thank you

Amogha

[Romain LE DONGE]

Hello,

I'm not a Data Scientist expert and it depends on your data and total dataset size, but I usually use the following split:

• Training dataset: 60% (70% if I miss data): data to be used to train the network
• Validation dataset: 20% (10% if I miss data): data to be validated during training process (used to plot train/validation curves)
• Test dataset: 20%: data to test the final performances of the network (never used during training or network topology research)

Hope this can help !

Romain