RuntimeError: You set the target layer to an activation layer

[Mayur Divate]

Hi,

I trying to implement DeepLIFT to interpret CNN model.
However, I am not able get through following error.

   target_layer_idx=-1)
  File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 113, in get_target_contribs_func
    return self._get_func(*args, func_type=FuncType.contribs, **kwargs)
  File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 234, in _get_func
    **kwargs) 
  File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 47, in _get_func
    self._set_scoring_mode_for_target_layer(target_layer)
  File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 133, in _set_scoring_mode_for_target_layer
    +" classification)")
RuntimeError: You set the target layer to an activation layer, which is unusual so I am throwing an error - did you mean to set the target layer to the layer *before* the activation layer instead? (recommended for  classification)

Process finished with exit code 1

My activation layer is softmax layer.
I was wondering whether DeepLIFT supports multi-class classification model?

Thanks,
Mayur

[Av Shrikumar]

Hi Mayur, It's throwing this error because in accordance with Section 3.6 ("Choice of Target Layer") of the DeepLIFT paper (https://arxiv.org/pdf/1704.02685.pdf), it is recommended that you set your target layer to the layer *prior* to the final nonlinearity to avoid artifacts caused by the saturating nature of a sigmoid/softmax. So if you set your target_layer_idx to -2, you should be fine. Does that make sense? - Avanti

…

On Thu, 24 Oct 2019 at 22:09, Mayur Divate ***@***.***> wrote: Hi, I trying to implement DeepLIFT to interpret CNN model. However, I am not able get through following error. target_layer_idx=-1) File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 113, in get_target_contribs_func return self._get_func(*args, func_type=FuncType.contribs, **kwargs) File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 234, in _get_func **kwargs) File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 47, in _get_func self._set_scoring_mode_for_target_layer(target_layer) File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/models.py", line 133, in _set_scoring_mode_for_target_layer +" classification)") RuntimeError: You set the target layer to an activation layer, which is unusual so I am throwing an error - did you mean to set the target layer to the layer *before* the activation layer instead? (recommended for classification) Process finished with exit code 1 My activation layer is softmax layer. I was wondering whether DeepLIFT supports multi-class classification model? Thanks, Mayur — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#90?email_source=notifications&email_token=AARSFBQRB4J44JEP5ZZTZQTQQJ5RPA5CNFSM4JE6TJV2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HUJFNDQ>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AARSFBTLXKA6ICBUCG5VFD3QQJ5RPANCNFSM4JE6TJVQ> .

[Av Shrikumar]

What are you supplying for the argument input_data_list? It needs to be a list of numpy arrays, one for each input mode.

On Thu, Oct 24, 2019 at 10:40 PM Mayur Divate ***@***.***> wrote: If I do so it throws new error. File "/Users/n10337547/PycharmProjects/Bitbucket/ModelInterpretation/venv/lib/python3.6/site-packages/deeplift/util.py", line 127, in run_function_in_batches while i < len(input_data_list[0]): TypeError: object of type 'Preprocessor' has no len() Mayur — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#90?email_source=notifications&email_token=AARSFBQASSQGENUFN6AF4A3QQKBD5A5CNFSM4JE6TJV2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECHHZ7Y#issuecomment-546209023>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AARSFBRSZMHF7ZFULDS5ZHTQQKBD5ANCNFSM4JE6TJVQ> .

-- Sent from my phone, please excuse brevity/typos

[Mayur Divate]

Sorry, I just realized that.
Wait I will try that.
Mayur

[Mayur Divate]

Hey, it worked!
Thanks,
Mayur