Error when saving TensorFlowModelDataset as partition

[anabelchuinard]

Description

Can't save TensorFlowModelDataset objects as partition.

Context

I am dealing with a project where I have to train several models concurrently. I started writing my code using PartitionedDataset where each partition corresponds to the data relative to one training. When I am trying to save the resulting tensorflow models as a partition, I get an error. I wonder is this has to do with the fact that those inherit from the AbstractVersionedDataset instead of the AbstractDataset. And if yes, I am interested to know if there is any workaround for batch saving those.

This is the instance of my catalog corresponding to the models I want to save:

tensorflow_models:
  type: PartitionedDataset
  path: data/derived/ML/models
  filename_suffix: ".hdf5"
  dataset:
    type: kedro.extras.datasets.tensorflow.TensorFlowModelDataset

Note: Saving one model (not as partition) works.

Steps to Reproduce

1. Generate a bunch of trained models
2. Try to save them in a partition as TensorFlowModelDataset objects

Expected Result

Should save one .hdf5 file per partition with the name of the file being the associate dictionary key.

Actual Result

Getting this error:

DatasetError: Failed while saving data to data set PartitionedDataset(dataset_config={}, dataset_type=TensorFlowModelDataset,
path=...).
The first argument to `Layer.call` must always be passed.

Your Environment

• Kedro version used (pip show kedro or kedro -V): kedro, version 0.18.12
• Python version used (python -V): 3.9.16
• Operating system and version: Mac M2

[astrojuanlu]

Hi @anabelchuinard, thanks for opening this issue and sorry for the delay. It will take us some time but I'm labeling this issue so we don't lose track of it.

[merelcht]

Hi @anabelchuinard, do you still need help fixing this issue?

[anabelchuinard]

@merelcht I found a non-kedronic workaround for this but would love to know if there is now a kedronic way for batch-saving those models.

[merelcht]

Using the PartitionedDataset is definitely the recommended Kedro way for batch saving. I've done some digging and it seems that the following lines are causing issues for using the TensorFlowModelDataset with PartitionedDataset:

kedro-plugins/kedro-datasets/kedro_datasets/partitions/partitioned_dataset.py

Lines 313 to 314 in be99fec

	if callable(partition_data):
	partition_data = partition_data() # noqa: PLW2901