Something wrong with labeling in classifier

[ArtificialNotImbecile]

when I use this data set
in classification problem. The metric=accuracy was below 0.4 and the prediction in test data is about 0.5(random guess, however I use the same algorithm in python gives me about 0.6 accuracy), auc metric behaves consistent with python sklearn algorithms during run phase. But the prediction in the test data does not match the auc given by stacknet command line information.(Again, it looks like a random guess)

One possible problem is that the data use (-1,1) to encode the target class, which is not normal but since sklearn can handle this pretty well I really hope stacknet can do this as well!

[kaz-Anova]

Could you share the whole command you used, parameters file and if possible the dataset (if you did any changes to that one). In principle, it should work with -1/1, but there could be a bug.

[ArtificialNotImbecile]

params
command I used

And I made a mistake, the prediction accuracy and auc score is normal in the test data, but command line information about accuracy is wrong. The problem exist when running the original data. Thanks!

p.s. I change the params.txt file a lot to test whether there is actually a problem. When use logistic regression the prediction accuracy is about 0.4, but in SklearnRandomforest is close to 0.5 and sometimes beyond 0.5(But notice that logistic regression actually performs better than randomforest)
@kaz-Anova