Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
AutoOut: Automated Outlier Detection and Treatment Tool (github.com/matelabs)
31 points by kailashahirwar1 on Aug 31, 2019 | hide | past | favorite | 6 comments


Dangerously opaque tool. You should tell users why a row is an outlier, and you shoul document the methods you use (even if it’s just links to sklearn). Also would be more valuable as a cli app


https://github.com/MateLabs/AutoOut/blob/9c3b1f4195aa96e829c...

Looks like it applies multiple algos:

{"model": "zscore", "params": {"threshold":3.5}}, {"model": "DBSCAN", "params": {}}, {"model": "OPTICS", "params": {}}, {"model": "IsolationForest", "params": {}}, {"model": "EllipticEnvelope", "params": {}}, {"model": "OneClassSVM", "params": {}}, {"model": "LocalOutlierFactor", "params": {}},


"AutoOut is an automated outlier detection and treatment tool that allows you to get better models with even better accuracy without writing a single line of code."

Is this ironic? If not, shouldn't you somehow describe your method (including references) for how to detect and "treat" outliers?


Seems like a simple tool that takes data files and tells about outliers. Why does it have to be used with a web browser? If one runs it locally, it probably should primarily have a command line interface.


So I just glanced over the code, or more specifically https://github.com/MateLabs/AutoOut/blob/master/app/outlier_... - and from what I can see, this is some kind of ensemble (learning) method which votes out the predicted outliers?


Agree with the comments below. I understand that the tool might be created with statistics-illiterate people in mind as users, but as now it is on the first page of Hacker News page we would expect a bit more details.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: