Common Output Errors

Common Output Errors or Warnings

• "BacktrackingSearcher.minimize: stepSize underflow"

1. This can be caused by collinearity issue in the logistic regression component. Please make sure you remove features that gets activated in all data points within each HMM(KC) (except the basic intercept/bias feature), and features that are perfectly correlated. For highly correlated features, we also suggest to remove them, but we haven't done extensive experiments on it. Please contact us if you meet further problems.
2. This can also be caused by the result of the linear combination of feature values and weights (which is put into logistic regression function) being too big/small resulting in almost 1/0 probabilities. You can try changing the initial feature weight bounds from default value 10 to small value (e.g., 0.1) to address this problem.