At the end of the day, I just want to do some machine learning

[Jrinne]

@danp

jlittleton: Thanks for running that test. You are correct that the datauniverse endpoint is returning weekly data. I didnt take screenshots of my results, so I dont know exactly what I tested. I know I saw the values (FRank results) change slightly day to day, but I cant reproduce that now. The lead API developer is out next week, but I will talk with him when he gets back to see what the options are regarding returning daily data from the API.

P123,

This will give me what i need and I would appreciate if you could put progress on this into the forum, P123.

And my apologies:

1. in the past there has been a competition of ideas as we did not have DataMiner and some of what we wanted to do had to be done within P123 or with limited downloads. We had to ask permission. People wanted different things and wanted their ideas addressed first.

2. Marc, at least but maybe not just he, was hostile calling everything I did F%$king Physics envy. I did respond in kind and probably would again.

Recently, however, there has been the ability to do what anyone wants with the downloads. I have tried to spark interest in machine learning and thought P123 wanted to be a machine learning platform ultimately. Bringing regression, support vector machines, random forests, XGBoost and neural nets to the platform, I had understood.

All while both P1213 and its members criticize these methods. Yuval recently had some comments on regression. Maybe he can clarify his support for P123 moving ahead with machine learning.

And I recently receive a message that machine learning may not be a suitable topic for P123.

Also I think people do not understand me. That is on me. I work hard to try to write well multiple revisions and probably too much length in an attempt to be clear.

But also I do not understand what the forum wants and I will work on that. Shorter post for one.

But I simply do not understand why someone will post a paper but be totally disinterested in the methods used in the paper.

Or ask a quesiton about why removing one factor at a time works (recursive elimination). But not want to know that coliinearly may play a role.

Or to be shown that regurlarzation is one very effective and usable way to eliminate overfitting and not be interested in how it could be implemented…

Or why MOD() is such a great technique but random seed is really, really complex and not worth considering as a feature.

Anyway, when Marc was here I was happy to respond in kind without the c%*ss words myself. I left it to him without really minding that he did that. I found it kind of charming. I miss him.

While I do miss Marc I had to question his idea of adding randomness to P123 to avoid overfitting. The hostility to machine learning was extreme and I find that things may not really have changed that much at P123—truth be told.

Put simply, if P123 has any desire to be a machine learning platform, I just do not get it.

I will try to regroup, focusing on my DataMiner download—especially if I can use them in a system at the time of rebalance, finally. I wanted that for a very long time and I appreciate finally getting it.

And if P123 ever become a place that does some machine learning and people want to support that in the forum i will try to contribute and understand how to best do that.

But below is very easy machine learning method (made necessary by the downloads available). Totally funded and out-of-sample. No cherry-picking. These are the only ports I have run since these start dates Errors and changes included (was 30 stocks now 15) universe changes, minor ranking method changes, Thursday went a while with no new stock buys because I removed a lot of cash.

For some reason I find this valuable and I want to support it. While the forum is hostile at times I do learn from posting. I really would not have developed the below system without posting for whatever—largely unexplainable—reason. I have shared basically all my methods in the forum at some point.

Anyway, I want P123 to do well. I will help with machine learning to the extent people want to discuss it in forum and will try to learn what people want…

Did I understand that P123 will be doing support-vector-machines? Maybe some backpropagation when t we have neural nets. Will some of this be an okay topic then?

Anyway, why I use machine learning and find value in it. Plus, I am getting better early results with a random forest:.I do want P123 to succeed with machine learning and keep providing DataMiner downloads with methods like random forest and XGBoost in mind. I think I will get that eventually and I can do it on my own then.

BTW, I know I may not be doing it right but no use of the optimizer for this.

Jim