Is there some demo about Uses Predict API that with multi params? what if tool occur error When I use in my model

[guirong guo]

as you Know I have test my TfRecord in local success, but use in what if too occur error as blow: 

Model inference failed: <_InactiveRpcError of RPC that terminated with: status = StatusCode.INVALID_ARGUMENT details = "input size does not match signature: 1!=4 len({input}) != len({dense_input,seq_input,sparse_ids_input,sparse_wgt_input}). 

[jwe...@google.com]

Unfortunately, none of the demos show use of the Predict API. The code for using this API only supports models where it expects its input as a serialized TF.Example proto which it deserializes and then parses features from (this is how the Predict API functioned originally, I believe now it allows multi-param passing).

To use WIT with a model like that, I suggest you write a custom prediction function in python that takes your examples from your TFRecord format and makes the correct API call to the model and returns the results, instead of having WIT try to use the Predict API. See instructions at https://pair-code.github.io/what-if-tool/learn/tutorials/custom-prediction/

Sorry for the inconvenience, I hope this helps.

[Joel Quesada Vallejo]

Hello! Do you plan to support this in the future? 

[jwe...@google.com]

We have no plans to add new features such as this. But you should be able to write a custom prediction function to make use of the Predict API as described previously in the thread.