what is the meaning of this '-1.#IND ' ?---actually i am getting t his when i am performing anfis and command window shows Start training ANFIS ... 1 2.82336 -1.#IND

Start training ANFIS ...

     1    2.82336    3.28319
     2    -1.#IND    -1.#IND

Designated epoch number reached --> ANFIS training completed at epoch 2.

ANFIS info: 
  Number of nodes: 910
  Number of linear parameters: 452
  Number of nonlinear parameters: 678
  Total number of parameters: 1130
  Number of training data pairs: 872
  Number of checking data pairs: 522
  Number of fuzzy rules: 113

Warning: number of data is smaller than number of modifiable parameters

Start training ANFIS ...

     1    2.82336    3.28319
     2    -1.#IND    -1.#IND

what is the meaning of " -1.#IND "?

Answer 1

[Walter Roberson]

-1.#IND is a NaN printed out by one of the more common MS Windows libraries. https://stackoverflow.com/questions/347920/what-do-1-inf00-1-ind00-and-1-ind-mean

Notice that (for whatever reason) ANFIS thinks you have 452 linear parameters and 678 nonlinear parameters for a total of 1130 parameters, but that you have only 872 training data pairs. That leaves you with more parameters than data pairs: when you have more parameters than available data, it is not uncommon for algorithms to end up computing nan (especially if the algorithm uses some kind of matrix inverse.)

[monideepa]

ANFIS info: 
  Number of nodes: 830
  Number of linear parameters: 412
  Number of nonlinear parameters: 618
  Total number of parameters: 1030
  Number of training data pairs: 1043
  Number of checking data pairs: 0
  Number of fuzzy rules: 103

Start training ANFIS ...

     1    3.29328
     2    -1.#IND

Designated epoch number reached --> ANFIS training completed at epoch 2.

i am increase the training data and it is greater than the total no of parameter , but epoch error shows -1.#IND. please tel me what i am doing now? can i taking the epoch error 3.29328 and generate rule for checking the data ? and another thing is that i am getting this epoch error at epoch no 1 i.e after first run. is this acceptable that epoch 1 gives minimum error?