I'm using the neuralnet package to try and predict a sample of some continuous data, but for some reason it only produces one predicted value, which interestingly enough is not the mean.
Is there any common known cause for this? I've tried wildly/randomly varying the values of threshold, the number of nodes and the learningrate to no avail, as the predictions are still a constant.
Here's some output from the learning process
hidden: 40 thresh: 0.3 rep: 1/10 steps: 762 error: 0.39832 time: 5.65 secs
hidden: 40 thresh: 0.3 rep: 2/10 steps: 18 error: 0.33205 time: 0.21 secs
hidden: 40 thresh: 0.3 rep: 3/10 steps: 22 error: 0.33204 time: 0.18 secs
hidden: 40 thresh: 0.3 rep: 4/10 steps: 22 error: 0.33203 time: 0.19 secs
hidden: 40 thresh: 0.3 rep: 5/10 steps: 14 error: 0.33206 time: 0.11 secs
hidden: 40 thresh: 0.3 rep: 6/10 steps: 16 error: 0.33206 time: 0.14 secs
hidden: 40 thresh: 0.3 rep: 7/10 steps: 19 error: 0.33204 time: 0.14 secs
hidden: 40 thresh: 0.3 rep: 8/10 steps: 22 error: 0.33207 time: 0.19 secs
hidden: 40 thresh: 0.3 rep: 9/10 steps: 1000 min thresh: 1.054522176
1300 error: 0.3486 time: 11.05 secs
hidden: 40 thresh: 0.3 rep: 10/10 steps: 23 error: 0.33206 time: 0.22 secs
Has anyone experienced or fixed a similar problem?
