uDreyfus1 {NNbenchmark}

Dataset uDreyfus1

Description

An univariate dataset (x, y) of class data.frame and dim 51 x 2 to be fitted by a neural network with 3 hidden neurons (10 parameters). This dataset was used to teach neural networks at ESPCI from 1991 to 2013. It usually appeared in the very first slides. This is a combination of 3 pure tanh() functions without noise. The Jacobian matrix is singular at the target parameter values and many algorithms could fail.

References

Dreyfus, G., ESPCI https://www.neurones.espci.fr

Examples

ht(uDreyfus1)
plot(uDreyfus1)