0

Good evening and nice to meet you all. I've been asked for a project to use an autoencoder for anomaly detection purposes. The dataset (synthetic, created by me) consists of 9 fictitious sensor readings.

The problem is that the request is to have 90 neurons in the input layer of the autoencoder, so what I've been actually asked to do is to collect vectors of 10 samples per each sensor (10*9=90) in order to have a 90-dimensional feature vector as input to the net.

Do you have some hints? Thank you

Mohsen
  • 4,536
  • 2
  • 27
  • 49
Adam
  • 62
  • 6
  • The problem is: he asked me to take a window of 10 samples per each sensor, stack it in a vector and feed it to a 90-Dim input layer. Does it make any sense at all? – Adam Nov 13 '18 at 19:05

1 Answers1

0

Sample ten times your synthetic dataset.

The goal here is to have a mini time-serie, I suppose, so you want to have a 10-sample serie for each of your 9 inputs.

Matthieu Brucher
  • 21,634
  • 7
  • 38
  • 62