Target from LabelBinarizer detected as `task type: multilabel-indicator`

[smith558]

When using the model below for multi-class classification and passing y targets from LabelBinarizer, these get detected as task type: multilabel-indicator instead of task type: multiclass.

def get_model(hidden_layer_sizes, meta, compile_kwargs):
    model = keras.Sequential()
    inp = keras.layers.Input(shape=(meta["n_features_in_"]))
    model.add(inp)
    for hidden_layer_size in hidden_layer_sizes:
        layer = keras.layers.Dense(hidden_layer_size, activation="relu")
        model.add(layer)
    if meta["target_type_"] == "binary":
        n_output_units = 1
        output_activation = "sigmoid"
        loss = "binary_crossentropy"
    elif meta["target_type_"] == "multiclass":
        n_output_units = meta["n_classes_"]
        output_activation = "softmax"
        loss = "sparse_categorical_crossentropy"
    else:
        raise NotImplementedError(f"Unsupported task type: {meta['target_type_']}")
    out = keras.layers.Dense(n_output_units, activation=output_activation)
    model.add(out)
    model.compile(loss=loss, optimizer=compile_kwargs["optimizer"])
    return model

from scikeras.wrappers import KerasClassifier

pipeline = make_pipeline(
    SimpleImputer(strategy='constant', fill_value='None'),
    OneHotEncoder(sparse_output=True, handle_unknown='ignore'),
    StandardScaler(copy=False, with_mean=False),
    VarianceThreshold(threshold=0.05),
    KerasClassifier(
        model=get_model,
        hidden_layer_sizes=(100,),
        optimizer='adam',
        epochs=5,
        batch_size=15,
        verbose=1,
        random_state=42,
        warm_start=True,
        metrics='accuracy'
    ),
    verbose=True
)

label_binarizer = LabelBinarizer(sparse_output=False)
label_binarizer.fit(y_train['attack_type'])
y_train_encoded = label_binarizer.transform(y_train['attack_type'])

label_binarizer.fit(y_test['attack_type'])
y_test_encoded = label_binarizer.transform(y_test['attack_type'])

pipeline.fit(X_train, y_train_encoded)

throws

Cell In[18], line 19, in get_multiclass_network(hidden_layer_sizes, meta, compile_kwargs)
     17     loss = "sparse_categorical_crossentropy"
     18 else:
---> 19     raise NotImplementedError(f"Unsupported task type: {meta['target_type_']}")
     20 out = keras.layers.Dense(n_output_units, activation=output_activation)
     21 model.add(out)

NotImplementedError: Unsupported task type: multilabel-indicator

[adriangb]

Can you include some sample data (not real data, it can be 1-2 rows of made up data with the same shape / types as your data) so that I can build a test notebook to try this in?