Assign requires shapes of both tensors to match about networkml

[scottkelso]

Correct, feel free to close,

[cglewis]

I changed the rnn size in DeviceClassifier/OneLayer/opts/config.json so that SoS and OneLayer would match which might be the problem:
cc17326#diff-cc727211c401f3b2b1a0596ad1ea4e55L37

Did you also train against this commit, or a prior commit?

If you change the value back to 128 does the error go away?

As for the warning about minimum groups, yes that means you don't have enough data samples in those buckets (classes).

[scottkelso]

I replace the config.json and label_assignments.json file each time so every test I have done has been with the same config file (where rnn=128). I think the error must have just been in a select few commits...

I have since ran training and evals on 3e5a52d (latest), cc1f326 and 1fd5e3e and have discovered that the only ones that are throwing that error, are those of commit cc1f326 with rnn=128, irrespective of which commit was used to train. See results below.

Train on 3e5a52d (latest)

          rnn=128  rnn=100
3e5a52d : PASS     PASS
cc17326 : ERROR    PASS
1fd5e3e : PASS     PASS

Train on cc17326 (18th July)

          rnn=128  rnn=100
3e5a52d : PASS     PASS
cc17326 : ERROR    PASS
1fd5e3e : PASS     PASS

Train on 1fd5e3e (17th July)

          rnn=128  rnn=100
3e5a52d : PASS     PASS
cc17326 : ERROR    PASS
1fd5e3e : PASS     PASS

Apologies if much time was wasted on this issue. Thanks!

[cglewis]

So it's working in the latest on Master now? Or am I misunderstanding?