The algorithm does not work after iteration 1?
Closed this issue · 4 comments
wangg12 commented
This is the output of one training process. Here the first iteration goes well, while after that strange things happened.
The dataset is a toy dataset produced by multi-variate normal random
with cov=50*I_50 and mean: ones(1, 50), ones(-1, 50).
With ada in R, the error can reach about 0.18.
Please help me find out what is going wrong. Thanks.
Iteration: 1, error: 0.321375, avg tree size: 63, num trees: 1
Iteration: 2, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 3, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 4, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 5, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 6, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 7, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 8, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 9, error: 0.5, avg tree size: nan, num trees: 0
Iteration: 10, error: 0.5, avg tree size: nan, num trees: 0
wangg12 commented
This problem happens only when the loss type is "exponential". You can reproduce it by setting it in the demo code. Maybe there are problems in the code for 'e'.
dmarcous commented
Hi, thanks for the catch! I'll try to look into it as soon as possible.
(Your are welcome to contribute a fix if you find one, I won't get to it before the weekend)