Algorithms details

[sevilaybayatli]

Hello,

Is vowpal_wabbit just used supervised algorithms? While I was reading this tutorial, I recognized that all algorithms are supervised. If I am wrong, please correct me? I need to use unsupervised learner and I am happy to use vowpal_wabbit.

Thanks million for your help.

[arielf]

Generally, you are right. Most of vw is supervised learning. In particular, there's no clustering support.

Exceptions:

• There are some matrix decomposition algorithms (--rank, --lrq, ... options) which are not strictly supervised.
• There's support for active learning (--active, --mellowness, --simulation, ... options) which is semi-supervised.
• There's indirect support for outlier/anomaly detection by real-time introspection of the streaming loss. See utl/vw-top-errors

[JohnLangford]

LDA is an unsupervised learning algorithm.

-John

On Mon, Aug 20, 2018 at 5:30 PM Ariel Faigon notifications@github.com wrote:

Generally, you are right. Most of vw is supervised learning. In particular, there's no clustering support.

Exceptions:

• There are some matrix decomposition algorithms (--rank, --lrq, ... options) which are not strictly supervised.
• There's support for active learning (--active, --mellowness, --simulation, ... options) which is semi-supervised.
• There's indirect support for outlier/anomaly detection by real-time introspection of the streaming loss. See utl/vw-top-errors

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/JohnLangford/vowpal_wabbit/issues/1572#issuecomment-414469932, or mute the thread https://github.com/notifications/unsubscribe-auth/AAE25puM_zuNghh2oo7rOueVSfOTZEL7ks5uSyp6gaJpZM4WDFiI .

[lokitoth]

@sevilaybayatli: Has your question been resolved?

If not, could you provide a bit more details about your specific scenario?

[jackgerrits]

Closing this issue as it seems like the question is answered. @sevilaybayatli if it is not, please go ahead and reopen it.