I'm working on a multiclass classification problem (12 classes), but I can't create the confusion matrix. What I'm trying to do is:
from sklearn.metrics import multilabel_confusion_matrix
pred = model_BiLSTM.predict(X_val)
y_unique = np.unique(y_val)
mcm =confusion_matrix(y_val, pred, labels = y_unique)
But as per the title, the returned error is: "ValueError: Classification metrics can't handle a mix of multiclass and continuous-multioutput targets", and the same happens when using confusion_matrix().
Instead, this is the output of
y_val, pred, y_val.shape, pred.shape
16/16 [==============================] - 0s 6ms/step
(array([11, 8, 3, 3, 9, 9, 9, 3, 3, 1, 3, 6, 6, 6, 2, 9, 1,
3, 10, 6, 9, 2, 9, 4, 3, 9, 6, 9, 3, 3, 3, 3, 8, 1,
9, 9, 9, 9, 2, 1, 9, 3, 9, 3, 9, 9, 2, 10, 3, 2, 9,
3, 8, 6, 8, 9, 9, 6, 6, 9, 8, 7, 9, 2, 9, 9, 2, 3,
9, 9, 2, 6, 3, 7, 11, 9, 9, 2, 11, 6, 7, 11, 6, 9, 2,
6, 9, 2, 9, 9, 3, 6, 9, 1, 11, 4, 4, 2, 6, 2, 9, 3,
10, 3, 2, 9, 9, 10, 6, 3, 9, 9, 6, 8, 6, 9, 4, 6, 5,
9, 6, 6, 3, 3, 3, 9, 6, 2, 9, 11, 6, 9, 3, 9, 6, 9,
2, 9, 3, 9, 6, 1, 6, 9, 9, 8, 3, 2, 9, 2, 8, 9, 9,
3, 10, 2, 4, 9, 9, 2, 8, 3, 6, 9, 9, 9, 6, 2, 2, 9,
7, 9, 3, 6, 7, 2, 9, 2, 9, 10, 9, 2, 9, 5, 7, 6, 5,
6, 3, 9, 7, 9, 3, 11, 5, 3, 3, 0, 8, 3, 9, 5, 9, 10,
9, 3, 3, 11, 2, 1, 8, 6, 2, 9, 3, 6, 2, 8, 9, 2, 8,
3, 3, 9, 6, 2, 1, 9, 9, 2, 2, 10, 9, 1, 6, 9, 9, 2,
9, 5, 6, 3, 9, 7, 6, 9, 9, 6, 3, 3, 2, 3, 9, 6, 9,
9, 9, 9, 9, 9, 9, 2, 9, 9, 9, 9, 3, 1, 6, 6, 6, 10,
9, 9, 9, 2, 9, 2, 3, 9, 9, 10, 6, 10, 6, 1, 1, 6, 1,
6, 3, 4, 6, 1, 11, 3, 9, 9, 4, 9, 9, 5, 8, 3, 5, 9,
6, 9, 9, 9, 3, 2, 5, 1, 9, 6, 1, 3, 9, 5, 3, 9, 6,
7, 2, 9, 9, 3, 5, 9, 6, 6, 3, 1, 9, 3, 9, 7, 9, 9,
7, 9, 10, 2, 9, 4, 9, 9, 9, 3, 9, 5, 3, 2, 9, 3, 8,
2, 9, 3, 11, 3, 3, 9, 2, 10, 9, 3, 9, 1, 9, 10, 5, 1,
2, 5, 9, 2, 3, 5, 9, 4, 8, 9, 5, 10, 3, 2, 6, 3, 6,
10, 11, 3, 6, 9, 3, 3, 0, 6, 7, 8, 9, 6, 5, 3, 1, 9,
2, 9, 9, 5, 2, 5, 3, 6, 11, 2, 9, 3, 6, 2, 9, 9, 3,
3, 3, 5, 6, 4, 4, 9, 2, 2, 5, 1, 0, 9, 4, 3, 3, 9,
9, 5, 2, 2, 2, 5, 7, 9, 3, 9, 9, 1, 2, 9, 6, 8, 2,
3, 4, 2, 3, 3, 2, 6, 5, 9, 5, 2, 2, 2, 9, 9, 6, 2,
9, 4, 9, 9, 2, 3, 8, 11, 9, 9], dtype=int32),
array([[5.7949889e-03, 2.5301890e-03, 5.9659913e-05, ..., 2.7534673e-03,
1.8798949e-03, 4.0977496e-01],
[2.1629781e-04, 1.0219574e-02, 1.2285617e-03, ..., 4.0498661e-04,
3.6948815e-04, 8.3618681e-04],
[8.1547890e-03, 1.1354284e-04, 1.3678521e-04, ..., 3.6535120e-01,
1.1546685e-03, 3.5349184e-03],
...,
[1.1976730e-03, 6.8558909e-02, 8.7605380e-03, ..., 1.7384565e-01,
5.4570078e-04, 2.0005915e-02],
[2.1097453e-02, 7.7744485e-03, 2.5690982e-01, ..., 5.4854238e-01,
3.9467164e-03, 1.6034273e-02],
[2.0812787e-03, 1.6885218e-05, 4.7070305e-05, ..., 4.3611538e-01,
3.6522493e-04, 1.4385413e-02]], dtype=float32),
(486,),
(486, 12))
If it helps, this is my model:
def build_BiLSTM_classifier(input_shape, classes):
input_layer = tfkl.Input(shape=input_shape, name='Input')
bilstm = tfkl.Bidirectional(tfkl.LSTM(128, return_sequences=True))(input_layer)
bilstm = tfkl.Bidirectional(tfkl.LSTM(128))(bilstm)
dropout = tfkl.Dropout(.5, seed=seed)(bilstm)
classifier = tfkl.Dense(128, activation='tanh')(dropout)
output_layer = tfkl.Dense(classes, activation='softmax')(classifier)
model = tfk.Model(inputs=input_layer, outputs=output_layer, name='model')
model.compile(loss=tfk.losses.SparseCategoricalCrossentropy(), optimizer=tfk.optimizers.Adam(), metrics='accuracy')
return model
What could I try to do?
Your pred array appears to contain class scores rather than predictions. Try pred.argmax(axis=1) instead.