Client timeout

[schollz]

Client timeout

[schollz]

Could you share your database?

[misisnik]

Yes for example this one
https://drive.google.com/file/d/18J75ydKUhA8hXd_zaH84w_GMLUdJlIAX/view?usp=sharing

[schollz]

Thanks @misisnik. The problem was that Linear SVM scales very very badly for your data, probably something to do with the decomposition of your data since you have one axis that is positive, magnetometer, while the rest are negative. I just added a timeout (6fba702) which will skip bad machine learning algorithms that take too long (>10 seconds).

As for accuracy, having one less machine learning algorithm won't make a huge difference. The classification scheme uses meta-learning, so it uses all the algorithms - and while the more the better, having one less won't hurt too much. Anyways, I see that your accuracy is pretty high when leaving that one out.