Modifications needed for multi-label classification

[shan18]

If I want to apply this model to a multi-class multi-label classification, what changes would be required for me to make on the Mask() layer? Can you give me any hint?
Will there be any other changes that I would have to make?

Thank You.

[Ryanglambert]

If you read/run the code as is, you will see this is already a multiclass classifier.

…

On Sat, Jan 19, 2019, 4:18 PM Shantanu Acharya ***@***.*** wrote: If I want to apply this model to a multi-class multi-label classification, what changes would be required for me to make on the Mask() layer? Can you give me any hint? Will there be any other changes that I would have to make? Thank You. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#88>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADmCPEiCcNqwekRdPmTHf5EqGvbwh59Yks5vE4ungaJpZM4aJVnS> .

[shan18]

@Ryanglambert I was asking for a multi-label classifier, where each image can contain multiple objects, so for example, the output of an image can be [0, 1, 1, 0, 0, 0, 1]. In the Mask layer at line 56, instead of argmax, there should be a different operation which would select the probabilities to be either 1 or 0 based on a threshold. Please correct me if I am wrong.

CapsNet-Keras/capsulelayers.py

Lines 47 to 62 in 923809b

	def call(self, inputs, **kwargs):
	if type(inputs) is list: # true label is provided with shape = [None, n_classes], i.e. one-hot code.
	assert len(inputs) == 2
	inputs, mask = inputs
	else: # if no true label, mask by the max length of capsules. Mainly used for prediction
	# compute lengths of capsules
	x = K.sqrt(K.sum(K.square(inputs), -1))
	# generate the mask which is a one-hot code.
	# mask.shape=[None, n_classes]=[None, num_capsule]
	mask = K.one_hot(indices=K.argmax(x, 1), num_classes=x.get_shape().as_list()[1])
	# inputs.shape=[None, num_capsule, dim_capsule]
	# mask.shape=[None, num_capsule]
	# masked.shape=[None, num_capsule * dim_capsule]
	masked = K.batch_flatten(inputs * K.expand_dims(mask, -1))
	return masked

And when I ran the code as it is on my dataset, the accuracy was not improving, it wasn't going above 0.5. So I just wanted to know, is there any place else except the Mask layer where the changes have to be made for a multi-label classifier?

[XifengGuo]

@shan18
By default, during training, the mask is provided by y, so you needn't change the line#56.
If the accuracy does not improve, you can track the training loss to see a NaN?

[shan18]

Okay, understood. Thank you for your help.