Example A Transformer-based recommendation system does not work when using Keras 3.4.1

[vladimirvlasevsky]

Issue Type

Bug

Source

source

Keras Version

Keras 3.4.1

Custom Code

No

OS Platform and Distribution

Windows 10

Python version

3.12.3

GPU model and memory

No response

Current Behavior?

I want to reproduce this example using Keras 3.4.1, but I get an error message. I noticed that this example works when using Keras 3.2.1.

Standalone code to reproduce the issue or tutorial link

The original code (https://github.com/keras-team/keras-io/blob/master/examples/structured_data/movielens_recommendations_transformers.py) is used, no changes.

Relevant log output

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[1], line 539
    536 train_dataset = get_dataset_from_csv("train_data.csv", shuffle=True, batch_size=265)
    538 # Fit the model with the training data.
--> 539 model.fit(train_dataset, epochs=5)
    541 # Read the test data.
    542 test_dataset = get_dataset_from_csv("test_data.csv", batch_size=265)

File ~\AppData\Local\Programs\Python\Python312\Lib\site-packages\keras\src\utils\traceback_utils.py:122, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    119     filtered_tb = _process_traceback_frames(e.__traceback__)
    120     # To get the full stack trace, call:
    121     # `keras.config.disable_traceback_filtering()`
--> 122     raise e.with_traceback(filtered_tb) from None
    123 finally:
    124     del filtered_tb

File ~\AppData\Local\Programs\Python\Python312\Lib\site-packages\keras\src\models\functional.py:244, in Functional._adjust_input_rank(self, flat_inputs)
    242             adjusted.append(ops.expand_dims(x, axis=-1))
    243             continue
--> 244     raise ValueError(
    245         f"Invalid input shape for input {x}. Expected shape "
    246         f"{ref_shape}, but input has incompatible shape {x.shape}"
    247     )
    248 # Add back metadata.
    249 for i in range(len(flat_inputs)):

ValueError: Exception encountered when calling Functional.call().

Invalid input shape for input Tensor("functional_1/Cast_1:0", shape=(None,), dtype=float32). Expected shape (None, 3), but input has incompatible shape (None,)

Arguments received by Functional.call():
  • inputs=OrderedDict({'user_id': 'tf.Tensor(shape=(None,), dtype=string)', 'sequence_movie_ids': 'tf.Tensor(shape=(None, None), dtype=string)', 'sequence_ratings': 'tf.Tensor(shape=(None, None), dtype=float32)', 'sex': 'tf.Tensor(shape=(None,), dtype=string)', 'age_group': 'tf.Tensor(shape=(None,), dtype=string)', 'occupation': 'tf.Tensor(shape=(None,), dtype=string)', 'target_movie_id': 'tf.Tensor(shape=(None,), dtype=string)'})
  • training=True
  • mask=OrderedDict({'user_id': 'None', 'sequence_movie_ids': 'None', 'sequence_ratings': 'None', 'sex': 'None', 'age_group': 'None', 'occupation': 'None', 'target_movie_id': 'None'})

[jsoung]

FWIW, this can be worked around by returning an OrderedDict from the method create_model_inputs().

def create_model_inputs():
    input_features = OrderedDict()

    input_features["user_id"] = keras.Input(
        name="user_id", shape=(1,), dtype="string"
    )
    input_features["sequence_movie_ids"] = keras.Input(
        name="sequence_movie_ids", shape=(sequence_length - 1,), dtype="string"
    )
    input_features["sequence_ratings"] = keras.Input(
        name="sequence_ratings", shape=(sequence_length - 1,), dtype=tf.float32
    )
    input_features["sex"] = keras.Input(name="sex", shape=(1,), dtype="string")
    input_features["age_group"] = keras.Input(
        name="age_group", shape=(1,), dtype="string"
    )
    input_features["occupation"] = keras.Input(
        name="occupation", shape=(1,), dtype="string"
    )
    input_features["target_movie_id"] = keras.Input(
        name="target_movie_id", shape=(1,), dtype="string"
    )

    return input_features