support for regression model

[samithaj]

to get a regression value from DNNLinearCombinedRegressor, can I do it with Predict SignatureDef or do I have to change it to Regression SignatureDef and do i need to make a custom signature_def_map when exporting

[MtDersvan]

@samithaj Predict SignatureDef is a very versatile signature, it accepts an arbitrary amount of input and output tensors for inference, while Regression SignatureDef only 1 input and 1 output (more here SignatureDefs), so you can definetly use it. Also, you need to keep in mind that you can map any tensor, but you need to know/specify their names. The generic ones are 'inputs' and 'outputs'.

[samithaj]

Is it possible to get a regression output from Predict SignatureDef or do i have to use Regression SignatureDef

seems like that need some hazzle , and Predict SignatureDef seems easier

When using the Classify and Regress APIs, TensorFlow Serving feeds serialized tf.Examples to the graph, so your serving_input_receiver_fn() should include a tf.parse_example() Op. When using the generic Predict API, however, TensorFlow Serving feeds raw feature data to the graph, so a passthrough serving_input_receiver_fn() should be used.

https://www.tensorflow.org/versions/r1.3/get_started/export

[samithaj]

It just works , i did't see it because pycharm debugger doesn't show tensors correctly ,when i print the result_future.result() the regression prediction is there But seems like your way of exporting only exports real_valued_columns you can check it in the client, if you change any _bytes_feature 's in feature_dict's name 'education': 'gender': to any other ex: 'educationGG': it doesn't gave any error I'm trying to solve this by changing all functions from tf.contrib to new canned functions

[MtDersvan]

@samithaj I updated the tutorial to work with r1.3. Please, check if it works for you, and if you have any problems I'd be willing to help.

[samithaj]

Still it does not export the embedding_column types ,i think it only exports the numeric_column types so my predictions are totally different from vs predictions performed inside python (est.predict) You can see clearly this by when you comment a _float_feature in wide_and_deep_client.py it gives an error but when you comment a _bytes_feature it doesn't give a error I found a cloud ml example here and it uses a custom serving_input_fn to do this but it's quring it with gcloud ml-engine local predict so not sure about how to get it working it for python client

[MtDersvan]

I'm not sure if I can help, but if you'll post code examples of what you are trying to do (what doesn't work), and an example of what you would like to achieve - I can try to extrapolate.