V-Training Service is active now??

[Nabeshin]

I uploaded files to the V-Training service but it seems not running....
Currently Task No.6(2020/12/29) is mine,
but task No.1 which was uploaded in 2020/12/24 still remaining.

Anyone can help please?

http://v-training.m5stack.com

Thank you.

[Zontex]

@nabeshin said in V-Training Service is active now??:

http://v-training.m5stack.com

Hi, we are aware of that and working to fix it ASAP.

[Nabeshin]

Thank you.
This is not serious at all.
I will try it again in next year.

Have nice holidays!!!

[Nabeshin]

@Zontex
Now I can upload files. But Training always fails with this error.

CONTENT: Unexpected Error Found During Triaining, A target array with shape (128, 6) was passed for an output of shape (None, 7) while using as loss categorical_crossentropy. This loss expects targets to have the same shape as the output.

I uploaded following data.

• 6 classes training data
• Each class has 90 files(85 files in train folder. 5 + 1 files in each valid folder)
  • At first, using the result folder created by UnitV_boot_v1220.py, I got Error: Lake of Enough Valid Dataset, Only 30 pictures found, but you need 35 in total.
  • Maybe Lake of is typo of Lack of I think.
  • So I moved 1 file manually from train folder to valid folder for each class.(Then I made 5 + 1 files in each valid folder.(Total 36 in valid folder)
  • Then that goes fine.
• But now I got that error about categorical_crossentropy.
  • I tried several cases and got same kind of error.
  • For 6 class case: A target array with shape (128, 6) was passed for an output of shape (None, 7)
  • For 4 class case: A target array with shape (64, 4) was passed for an output of shape (None, 5)

I'm not sure but I think it's because,

• At model.add(Dense(X)) script, X should be the same number of uploaded class folders of each request.
• But maybe it is over counted from exact one in your script.
  • https://github.com/keras-team/keras/issues/3237

Do you have any idea for fix it?

Regards.

[Nabeshin]

@Zontex
Hi, I tried again now but It seems like still having problems...

Can I build same environment in my computer?
What should I install for this machine learning?

If you can point the GitHub URL, Please let me help.

Regards.

[Nabeshin]

Hi, Now I got one solution by myself.
We can build a private environment of K210 machine learning on Goggle colaboratory.
Of course we can share.

I wrote an article here.
This can be used as an another way of V-Training for M5StickV and M5UnitV.

Qiita

• https://qiita.com/Nabeshin/items/d526a94b8f3be85bbb05

Goggle colaboratory

• https://colab.research.google.com/drive/1UdQV5nQMQy1ckJP0qdS3kS7sW6sEQyzs#scrollTo=prOt_VofnYCo

Thank you.