The MetricAnalyzer is a free tool to find good strategies

Hello, at this point I would like to introduce you to my metrics analyzer. This tool can help you create more successful strategies with StrategyQuantX. This is the first version of the metrics analyzer. I will do additional update next week so that the installer will install a version and a complete demo will also be installed. This makes it even easier to get to know everything. The metrics analyzer is freeware, you can use the tool freely. The metric analyzer is a free tool that offers some simple features for analyzing the strategy generator. I have found that the most important thing when generating strategies is the generator. If it’s good, then the strategies will work. The metrics analyzer is not intended to compete with StrategyQuantX. I hope that the SQX team will integrate some features of the analyzer into a similar improved version in one of the next versions of SQX. I don’t have that much time to realize all possible ideas. The tool should also serve as a suggestion for what else you could do.

Next week I will integrate a demo into the installer so that you can get to know how the metric analyzer works even more easily.
I am also aware that the metrics analyzer may over-optimize too much. I already have an idea of ​​what I’m going to integrate. If you want, you can try out version 0.1.4.3 (without demo mode). I have attached the current documentation here. You can download the Metric Analyzer from my gmx-webspace.

https://c.gmx.net/@329881123612003410/AXjh2A75Rm-xTLKwpAVWSA

thomas

I updated the Metrikanalyser.

1) I added a Demomode installation for easier understanding this all

2) I updated the Installer

thomas

I am working on the next update for the Metrikanalyser, integrating a small robustness test. I will test the top 100 Metrikrules with unseen data, allowing you to review the results and decide whether to adopt the discovered rules.

Currently, there are only 77 strategies. I need to repeat this process with 10,000 or more strategies. Each set of unseen data constitutes 10% of all data, allocating 50% for generation and 50% for testing with unseen data.

I plan to update the Metrikanalyser with this feature in the coming days.

For the foreseeable future, I do not intend to update the Metrikanalyser. Instead, I will familiarize myself with Weka and explore various learning models. I believe this approach is more efficient than trying to reinvent the wheel.

thomas

Hey Tnickel!

I am doing teh same. With the worflows and the metrics too. Good Luck

Weka, now there is something I had never heard of before, highlighted the word and did a search internet and CoPilot in Edge lit up. Feel like a kid on a treasure hunt picking up clues-thanks.

I am happy with weka, it is a very good tool.I hope I could motivate many people to work with it.

I implemented the RandomForest Learner in the Metrikanalyser.

The Results are very interesting. I am not an KI Expert. I am searching for people who have knowledge in KI.

With random forest learner overfitting can occour too. I am experimenting at the moment. I worte a documentation about my research. I think I will release the documentation in the next days.

My first interim conclusion is the following. It’s not that easy. You just start up the Weka Learner and the AI ​​learns everything, but it’s not that simple.

My tip: we all have to work together otherwise we won’t succeed. You often find people here in the forum who know something but only pass on their knowledge if they get the same knowledge from others.

Unfortunately, that’s not how it works. We’ll be stuck trying to improve things for another 100 years. That’s only possible if everyone gets involved and shares their knowledge.

I’ll be sharing my AI research with Weka in the next few weeks.

https://en.wikipedia.org/wiki/Weka_(software)

It is a long way to find profitable strategies.

If we find a way, it will be with AI. I have investigated many things with Weka. But I will really start with Analysis21. Here I will automate many things. I am excited to see what the SQX team will incorporate into Strategyquant in the future in terms of AI. The Random Forest is not that easy, as I have found. You learn, but here too it depends on what you do. I have not been that successful with it yet. But you do learn. I have attached my analysis 20 times. I hope that this will motivate some more people to get to grips with Weka. The good thing is that Weka is written in Java and offers a lot of support if you want to implement the whole thing yourself. I am in the process of integrating parts into the toolbox and the metric analyzer.

It’s going to get very exciting now. I’ve updated the metric analyzer with Weka learner and Weka classifier. I’m trying to filter strategies using AI methods. For example, I generate M5 strategies for 4 months and do a final test of 4 weeks. Then I learn these strategies with the Weka Random Forest Learner. In the next step, I move the whole thing 4 weeks into the future. I take the learned Random Forest Learner from the past and apply it to the current period. This means that I filter the currently generated strategies using the metrics from 4 weeks ago. I hope that this approach works very well. An initial test is quite promising. I hope it works with the strategies and on the demo account and real account. There is a new update for the metric analyzer and of course new documentation. If all of this is too complicated for you, you can use the correlator, with which you can determine the most important attributes. The great feature. The whole thing isn’t completely finished yet. I would still like to publish the update.

I added the docu

I attached Analysis 21 and 22.