"No training configuration" for code that should not have one

[junyizhao04]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 20.02
• TensorFlow version and how it was installed (source or binary): 2.6.5
• TensorFlow-Addons version and how it was installed (source or binary): yes, by condo
• Python version: 3.7
• Is GPU used? (yes/no): yes

Describe the bug

I tried to fill in the spaces of the code provided in the tutorial with single model written here:

https://calamari-ocr.readthedocs.io/en/latest/doc.predicting.html

INFO 2023-05-18 23:47:16,968 calamari_ocr.ocr.savedmodel.sa: Checkpoint version 5 is up-to-date.
INFO 2023-05-18 23:47:17,012 tfaip.device.device_config: Setting up device config DeviceConfigParams(gpus=None, gpu_auto_tune=False, gpu_memory=None, soft_device_placement=True, dist_strategy=<DistributionStrategy.DEFAULT: 'default'>)
WARNING 2023-05-18 23:47:18,433 tensorflow: No training configuration found in the save file, so the model was not compiled. Compile it manually.
WARNING 2023-05-18 23:47:18,433 tensorflow: No training configuration found in the save file, so the model was not compiled. Compile it manually.

All files used here are from calamari files on Github.

[andbue]

Thanks for reporting! Just to be sure: this is only about the unnecessary warning, prediction works as intended?

[junyizhao04]

Thanks for reporting! Just to be sure: this is only about the unnecessary warning, prediction works as intended?

It's very weird: when you call python *.py, it does not work and run into some endless loop. but when calling python then doing import *, it works.

[andbue]

This is not related to the warning. You need a if __name__ == "__main__" block in your script, then calling it will work.

[junyizhao04]

Thank you! one more question: I saw previous works on Calamari OCR used Predictor.network, but I cannot call it today. Is it retired? If so, where can I fetch the network?

[andbue]

I'm not entirely sure at the moment, but I guess it's now at Predictor.model (https://tfaip.readthedocs.io/en/latest/tfaip.predict.html#module-tfaip.predict.predictor).

[bertsky]

It's at predictor.model now, indeed. For multi-model voters, it's only available via predictor._keras_model, which is an ad-hoc class JoinedModel, but that seems to inherit from keras.Model.

Can we close @junyizhao04?