Thread by @TRSTNGLRD

Alright, let’s talk Lorentzian Classification.

As most of you are aware, I’m fascinated with the study of Algorithmic Theory in Finance. I study a wide range of Algorithmic concepts, ranging from Chaos Theory & the Core Pricing Mechanism to Risk-Management & individual Trading… twitter.com/i/web/status/1…

Recently, I came across a video by @JustinDehorty, where he explains the idea of Lorentzian Classification and showcases an Indicator that utilizes this model as a Proof-of-Concept.

Below is a Summary of his video, followed by an overview of potential … twitter.com/i/web/status/1…

When you have an Indicator, such as an RSI or a Moving Average, there are actually a lot of issues that come from the data these derive - and most of you have actually probably analyzed this yourself. It’s a very well-known issue.

First, ask yourself how a Moving Average derives its data…

It averages out - simple. Hence, the name.

If you are looking at a dataset and you are to take the average of that dataset, then you are completely blocking out any peak event that takes place.

If you have an RSI that is showing ‘Oversold’, you ask yourself “What is the *relative* historical significance?”

Historically, when this RSI became oversold, what happened every other time the RSI became oversold?

If you have a Moving Average saying $150, price is currently at… twitter.com/i/web/status/1…

If price moves on average ~$10/Day, and one day, you get a move of $100 (10x Historical Average), there is something extreme that has taken place on that day.

If you use an indicator, it will not account for this data well, because it’s outside of its historical average. Got it?

This Phenomenon is what Justin describes as “Price-Time Distortion”, in reference to “Space-Time Distortion”.

He relates this concept to Gravity, where one event has far more significant mass than the others, causing a distortion in “Price-Time”. This throws Indicators off. twitter.com/i/web/status/1…

This is called “Euclidean Distance”, and below is a picture describing RSI, ADX, and CCI Data through this Distance. Visually, it appears as a tight cluster of thousands of datapoints.

This is basically how normal Algorithms interpret data; There is no structure to it.

This is where the “K-Nearest Neighbors ML Algorithm” comes into play.

First, let’s talk Machine Learning vs Deep Learning:

Machine Learning uses Machine Code & Structured Data to automatically adapt or predict w/ minimal human interaction.

Deep Learning uses Artificial Neural… twitter.com/i/web/status/1…

The “K-Nearest Neighbors Algorithm” is used as an extremely simple method of Data Classification / Prediction.

The kNN Algorithm actually specializes in large and significant datasets, which is what makes this concept so robust.

For those with a Computer Science background,… twitter.com/i/web/status/1…

As described above with the RSI example - if you are looking at an RSI, you are only interested in the precedence of it reaching an Oversold or Overbought value.

This is what kNN does.

Those extremes are classified as this datasets “Nearest Neighbors”, and it compares them.

This gets significantly harder to do manually when you include multiple indicators together…

Basically, you’d need to analyze the relative neighborhood of data for each individual indicator, and then compare all points together to manually find a pattern.

Basically impossible.

Well, on the other hand, this is that same data depicted through Lorentzian Distance. As you can see, there actually is structure to this data, and everything seems very organized and well-distributed.

When we lower the total points, focus in on that Red Dot where the mouse is hovering over.

The RSI value here was SIGNIFICANT compared to its other values (height), yet, it shares an equivalent value for both the CCI & ADX…

It structures by Relative Significance.

This cancels out this phenomenon of “Price-Time Distortion” entirely, as the weight or mass of one event is counteracted by the total weight or mass of the other events it’s taking into account (In this case: CCI, RSI, and ADX).

Well, okay - that’s cool I guess. But, this is all about Indicators… If you utilize ICT Concepts, you definitely don’t want anything to do with any indicators.

This is not what I find significant.

Let’s take a look at my teaser from a couple days ago…

As you can see, this algorithm attaches a value to each individual candlestick. You can consider this to be that value that’s related to Gravity, or the value of “Price-Time Distortion” that measures relative significance.

Let’s go deeper though, and see if we can visualize it. Below is the Beta-Version of this script, which attributes Colors to a Value.

Below are my settings; I want to make clear, I am *NOT* focused on the Indicator. I only care about the Classifications.

Now, what do you see?

Pay attention to the Values, Trend, and Market Structure.

Notice how *incredibly* well it is able to interpret a new Trend. Notice how incredibly perfect it is able to classify a Market-Structure Shift from a Retracement…

Within a Bearish Trend, the colors (values) will be… twitter.com/i/web/status/1…

This Algorithm is able to classify Candles to an absurd degree with extraordinary accuracy.

Now, let’s replace the “Features”.

Rather than Indicators, let’s use:
- Orderflow
- Volume
- Volatility
- Liquidity
- Points of Interest
- Interest Rates / COT
- Seasonality

What would that look like? Well, sadly, I can only Hypothesize.

However, what I have just explained to you is a major key in automating Smart-Money Concepts to achieve absurd accuracy.

You can take it a step further, and after implementing the kNN ML Algorithm, you then take… twitter.com/i/web/status/1…

Basically, I’m calling on *ALL* PineScript Coders that want to automate Smart-Money Concepts:

This Library is OPEN-SOURCED.

I advise that you heavily look into the applications of this Library, it is not to be neglected. This will improve algorithmic efficiency exponentially.

I’ve hit the thread limit, so this concludes my overview of Lorentzian Classification.

Please watch @JustinDehorty full video on the Model, and please check out his ML Library below to utilize it in future experimental scripts.

Link to Library: tradingview.com/script/ia5ozyM…

As you may know, I’m currently working towards various projects, such as an ICT-GPT that allows users to ask questions about the Core Content Mentorship.

If you are interested in my future works, please turn on notifications, share my page, and give me a follow!

Thank you!