Add export savedmodel to wide_deep #4041
Conversation
| --tag_set serve --signature_def="predict" \ | ||
| --input_examples='examples=[{"age":[46.], "education_num":[10.], "capital_gain":[7688.], "capital_loss":[0.], "hours_per_week":[38.]}, {"age":[24.], "education_num":[13.], "capital_gain":[0.], "capital_loss":[0.], "hours_per_week":[50.]}]' | ||
| ``` | ||
|
|
official/wide_deep/wide_deep.py
Outdated
|
|
||
| # Export the model | ||
| if flags.export_dir is not None: | ||
| wide_columns, deep_columns = build_model_columns() |
There was a problem hiding this comment.
Maybe separate out into a separate function to keep the main a little cleaner? Then just call the fn if export_dir is not None.
| "the model will be exported to this directory at the end of " | ||
| "training. See the README for more details and relevant links.", | ||
| metavar="<ED>" | ||
| ) |
There was a problem hiding this comment.
Should we provide a default directory if user provides none? Not sure if this feature is used frequently by users. If so, I think a default dir may be good.
There was a problem hiding this comment.
Considering that the functionality of SavedModel is currently limited to serving, I don't think this feature is frequently used.
|
Will keep in mind in the future -- thanks! |
* 'master' of https://github.com/tensorflow/models: (404 commits) official/mnist: Updates with the release of TensorFlow 1.8. Internal changes to slim and object detection (tensorflow#4100) catch cpuinfo ImportError (tensorflow#4138) [samples/core/get_starter/eager]: Update with API simplifications in 1.8 Add vincentvanhoucke as owner for research/marco Add marco subdirectory to README.md Create request.json Create jpeg2json.py Fix width / height. Create README.md Add new import paths to mnist_tpu.py (tensorflow#4133) fix model call (tensorflow#4132) Remove useless fields in resnet_model (tensorflow#4110) official/mnist: Add input_shape parameter to initial layer (tensorflow#4104) (tensorflow#4118) Increase minimum TF version for DEFINE_enum and rename variable mappings for change to RNN variable names. (tensorflow#4123) fix import errors Minor update the resnet runloop. (tensorflow#4113) Two new plot-generating scripts and changes in support of the ICLR poster. (tensorflow#4092) Add export savedmodel to wide_deep (tensorflow#4041) Fix 2 typos ...
|
Hi, guys! I have tried this command several times: saved_model_cli run --dir /My Directory/ --tag_set serve --signature_def="predict" --input_examples='examples=[{"age":[46.], "education_num":[10.], "capital_gain":[7688.], "capital_loss":[0.], "hours_per_week":[38.]}, {"age":[24.], "education_num":[13.], "capital_gain":[0.], "capital_loss":[0.], "hours_per_week":[50.]}]' But it is always wrong with an error NameError: name 'age' is not defined It seems none of keys make sense. and I read the source code, which do call build_model_columns() function and tf.feature_column.numeric_column('age') and so on. |
|
@vicdxxx I believe this is due to Windows treating single quotes as part of the input. Try running this:
For future questions, please create a new issue or create a Stackoverflow question (with the tensorflow tag) |
|
Hello, I need to install tensorflow-gpu 1.8 for a project. I have a Geforce GTX 1650ti graphics card. Can you show me a way to do this? @vicdxxx |
6 participants
Looks like wide deep is the only official model that doesn't currently export savedmodel. Here are the changes to export it.