using tf.linalg.det in custom loss function yields matrix inversion error about tensorflow HOT 5 OPEN

''' import tensorflow as tf
from keras.losses import Loss
class DeterminantZeroLoss(Loss):
def init(self, name="determinant_zero_loss"):
super().init(name=name)

def call(self, y_true, y_pred):
    try:
        # Calculate the determinant of the predicted matrix
        det = tf.linalg.det(y_pred)
    except tf.errors.InvalidArgumentError as e:
        # Handle the case where the matrix inversion fails
        # You can print the error message or take any other appropriate action
        print("Matrix inversion failed:", e)
        # Return a large loss value to indicate the failure
        return 1e6

    # Penalize when the determinant is close to zero
    loss = tf.abs(det)

    return loss

'''
hope it helps

from tensorflow.

except tf.errors.InvalidArgumentError as e:

Sadly that didn't catch the error, but I found out the error seems to only occur if I use
kernel_initializer=tf.keras.initializers.Zeros()
in one of the layers as well as the determinant=0 loss.

PS:

    return 1e6
# Penalize when the determinant is close to zero
loss = tf.abs(det)

You seem to have missunderstood that, when the matrix is not invertible the determinant should be zero, therefore the loss should be low, if the determinant is close to zero that is good and not penalized ;)

from tensorflow.

Hi @Sur3 ,

Could you able to submit a reproducible code snippet to debug this? Thanks!

from tensorflow.

This issue is stale because it has been open for 7 days with no activity. It will be closed if no further activity occurs. Thank you.

from tensorflow.

Sorry I didn't find time to produce a full code example yet, maybe next week.

from tensorflow.