The Forum and machine learning moderation


There is a thread about It mentions the advantage of subthreads. Since we do not have that feature yet I have started a new (sub)thread.

It is also true that P123 has dedicated an awful lot of time and resources to bringing machine learning to P123. P123 will be going ahead with that it seems. This continues to have potential but it is also true that you have not even started the beta and nothing is guaranteed. You want it to be a quality feature but also popular and bring in new members.

If you want AI to succeed at P123, I think you should look beyond changing the forum interface

Ideally, whoever developed machine learning at P123 will moderate a portion of the new forum. Respond to new feature suggestions and respond to concerns. Use the interaction to make improvements in the platform and grow P123 in general.

The guys and gals over at Kaggle are are a different type than you have encountered here on the forum. Maybe some of them will just use use the API and be pretty low maintenance: that will be a plus.

Still–if you want machine learning to bring in a lot of new members–the forum will probably need someone posting who believes in machine learning and is enthusiastic about explaining it. Be willing to make some changes in a focused–not too resource intensive–manner. Maybe even be able to discuss some of the resource limitations and ML methods on the fly.

Maybe things will be perfect and be “quite the thing” for the Kaggle crowd from the very first day. But there are some things that may still be unresolved such as the different ways that cross-validation can be done with a time-series. Plus, there are whole textbooks on hyperparameters. Not to mention parametric vs non-parametric methods, classifiers vs regressors, metrics such as recall vs precision vs balanced accuracy, Matthews correlation coefficient, f1( the favorite of many) etc. You may want to attract some undergraduates that do not have a degree in computer science or machine learning.

[color=firebrick]My prediction: If you even think about providing just accuracy as a metric (without the opportunity for serious discussion about metrics in the forum) many will immediately leave the P123 site (not just the forum) and never come back.[/color] There will be other concerns that I cannot predict (or even understand at this point).

[color=firebrick]And by the way, this will be a discussion that goes a little beyond the discussion of whether the information ratio (or standard deviation) has any value or not. I am alluding to Miro’s (Parker’) feature request for the information ratio back in 2014–clearly, it can take a while for implementation (P123 said they would implement the information ratio) and the discussion can be quite involved for just one metric no matter what your position on the metric is.[/color]

I get that this may not be of interest for the present forum. Perhaps, this post too should be marked as “SPAM” for now as has already been done with some posts about machine learning (according to my emails). But if that attitude continues at P123 the project is already dead, IMHO

BTW, you mentioned XGBoost will be available. It is an open-source program maintained by some of the people at the link below. Maybe you would like some of them to come to P123 and at least use the API. But who will answer their questions and maybe even take a feature suggestion or two seriously (for their benefit and for the growth of the machine learning portion of P123’s platform)?

You can take a look at the topics they are interested in here: XGBoost open-source platform forum.

Moderation for ML topics will be good, assuming the moderator understands the questions and the P123 moderator cares if AI succeeds or not.

It may not be too late to just cancel the project if you have not already thought about this. P123 has value just the way it is. If you do decide to cancel AI please delete this post along with any remaining posts of mine about ML.

Good luck whatever you end up doing.