Applying class_weight for Models with multiple "output"

[wosimidwa]

I am trying to train a Model using Keras and I have to apply class_weights to set the data ratio correctly. However, I do have 3 types of labels and when I try to train the model it gives the following error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
File f:\code\small\class_weight.py:2
      1 #%% Modeli Eğit
----> 2 history = model.fit(
      3     train_dataset,
      4     epochs=20,
      5     validation_data=val_dataset,
      6     class_weight=class_weights
      7 )

File ~\AppData\Roaming\Python\Python311\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\Roaming\Python\Python311\site-packages\keras\src\trainers\data_adapters\tf_dataset_adapter.py:128, in make_class_weight_map_fn.<locals>.class_weights_map_fn(*data)
    122 if tree.is_nested(y):
    123     raise ValueError(
    124         "`class_weight` is only supported for Models with a single "
    125         "output."
    126     )
--> 128 if y.shape.rank >= 2:
    129     y_classes = tf.__internal__.smart_cond.smart_cond(
    130         tf.shape(y)[-1] > 1,
    131         lambda: tf.argmax(y, axis=-1),
    132         lambda: tf.cast(tf.round(tf.squeeze(y, axis=-1)), tf.int32),
    133     )
    134 else:
    135     # Special casing for rank 1, where we can guarantee sparse encoding.

TypeError: '>=' not supported between instances of 'NoneType' and 'int'

here are some essential parts from my code:

#%% CLass Weight Calculation
total_samples = sum(class_counts.values())
current_ratios = {cls: count / total_samples for cls, count in class_counts.items()}
target_ratios = {0: 0.45, 1: 0.35, 2: 0.2}
class_weights = {cls: target_ratios[cls] / current_ratios[cls] for cls in class_counts}
print("Class Weights:", class_weights)

num_classes = 3
n_features = 24

model = Sequential()
model.add(LSTM(128, input_shape=(time_steps, n_features), return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(64, return_sequences=True))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(num_classes, activation='softmax'))

model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) #categorical_crossentropy -> one hot encoding
model.summary()

What should I do? I have researched the web but I couldn't find a way to apply class_weight properly for a model with multiple outputs.

[sonali-kumari1]

Hi @wosimidwa,

Thanks for reporting this issue. The code you provided seems to be a single output model with multi-class because you have only specified one loss and one metrics. To create a model with multiple outputs you would need to specify different losses and metrics for each output. However, I have tested your code with random data and it is working fine. Attaching gist for your reference.