I am using autoencoder for anomaly detection. I don't have any labels already and so its unsupervised. If I have categorical variables, I usually one hot encode before giving it to the model. I would like to know if we there is a possibility to give more weights to one particular categorical feature before giving it to the model.

Any help is much appreciated!!

  • $\begingroup$ @Did you find a solution to this issue. I need to do the same thing. $\endgroup$ – Deepak Saini Oct 11 '18 at 8:01
  • $\begingroup$ @DeepakSaini Not yet! $\endgroup$ – Ashwini Oct 13 '18 at 10:00

Do you mean giving more weights to a selected part of the input? If so, then, I believe this is possible with neural networks. Each neuron is assigned a random weights at the beginning of training. You can just initialize the first layer of the neural network model with some values representing the desired weights and then freeze those neurons.

An automated way of doing this is by using attention mechanism which automatically provides more weights (attention) to a portion of the input data.

|improve this answer|||||

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.