Thermometer Code

Thermometer coding is one approach to representing information that is to be presented to an artificial neural network. Thermometer coding is usually used to represent a quantitative variable. Imagine some variable of this type that varies in value from 0 to 10. To thermometer code this variable, one would turn on a sequence of input units, where the length of the sequence would equal the encoded value. For instance, to represent a value of "2" the first two input units would be activated; to represent a value of "8", the first eight input units would be activated, and so on. It is called a thermometer code, because the input units resemble a thermometer placed on its side, with the "height" of the "mercury" in the thermometer representing the value of the input variable. Dawson and Zimmerman (2003) used a thermometer code to represent distances when a network was trained on the Piagetian balance scale task.