Lorentzian Classification [Beta]

٢ مارس ٢٠٢٣

Summary of new Features:

Downsampling Toggle
Downsampling is a powerful technique because it helps to reduce the number of bars used for training the model, which in turn reduces the computational time required to train the model. This can be especially helpful when working with large datasets that contain a large number of bars. By selecting a subset of data points and ensuring a minimum spacing of 4 bars between neighbors, downsampling can help to add more diverse data to the model and reduce overfitting. Overfitting occurs when the model becomes too closely aligned with the training data and fails to generalize well to new data. By reducing overfitting, downsampling can improve the overall quality of the indicator and its ability to make accurate predictions.

Use Remote Fractals Toggle
There is now a new toggle for "Use Remote Fractals," which allows you to place additional emphasis on exotic fractals for training the model. By extending back to the beginning of the chart's history, this option can help you to add another perspective to your trading and potentially identify patterns that may not be apparent from the current chart. It may be less accurate than the default mode overall, but it can occasionally provide some impressive, unexpected insights.

New Reset Factor Field
There is now a new field with an adjustable “Reset Factor” metric. This factor can be thought of as being somewhat similar to the function of a dropout metric for deep learning-based techniques. It is a way for you to fully leverage the algorithm’s strength of giving you a diverse grouping of neighbors while still allowing you to control the rate at which more dissimilar neighbors are introduced to the neighborhood.

Customizable Colors
This is one of the most highly requested options for the main indicator. This version of the Lorentzian Classification script allows the user to customize the colors of the bars, the prediction text color, and the color of the signals. The default colors for these signals are now considered to be colorblind-safe.

٢ مارس ٢٠٢٣

ملاحظات الأخبار

• Updated version number to avoid confusion with the public indicator
  

١ يناير ٢٠٢٤

ملاحظات الأخبار

Summary of Latest Features:

Custom Source Inputs
With the new Custom Source Inputs, users gain the ability to seamlessly incorporate data series from a variety of indicators within the interface, transforming them into valuable ML feature series. This not only accelerates the computational process but also expands the scope of data analysis, allowing for a more extensive range of data points to be examined. This enhancement significantly increases the classification model's dimensionality, raising the total number of features from 5 to 8. This expansion is pivotal in providing a more nuanced and comprehensive analysis of market dynamics, offering deeper insights. Moreover, the process of testing and refining strategies becomes more fluid, as Custom Source Inputs facilitate quick and effortless modifications. This agility is crucial for promptly assessing the impact of any changes on strategy performance, fostering a dynamic and responsive approach to market analysis.

Price-Time Warping Reduction
A new toggle labeled "Reduce Price-Time Warping" has been introduced in this update. This toggle allows for a manual adjustment in the character of the distance metric used by the model. Specifically, it provides the option to increase the Euclidean character and reduce the Lorentzian character of the distance metric. One of the greatest strengths of the Lorentzian distance metric is that it is sensitive to the "price-time" warping effect often associated with significant world events, but this can sometimes produce suboptimal results in certain market conditions. By enabling an increase in the Euclidean character of the distance metric, this toggle can help mitigate such issues. This is particularly beneficial in periods of low volatility or in markets that are relatively insulated from major world events. While ideally, the selection between Euclidean and Lorentzian distance metrics would be a dynamic process, sensitive to the relative proximity of world events, for now, this has been implemented as a manual toggle. This new feature enhances the adaptability of the Lorentzian Classification Script, allowing for a more tailored alignment with ever-changing market conditions.

Kernel Regression Envelope
The new built-in Kernel Regression Envelope feature serves as a visual guide that graphically illustrates the potential range of price movements on the chart. This tool aids users in gaining a deeper understanding of market dynamics, thereby facilitating more informed trading decisions. Particularly for experienced traders, the envelope can serve as a valuable instrument for risk management, assisting in the identification of potential entry and exit points for strategic decision-making. Moreover, integrated into the envelope itself is an additional ML-driven classification layer, designed to assist in characterizing short-term mean-regressions. This feature leverages historical data in relation to the kernel-derived upper and lower envelope bounds, to assist in identifying bars that may signal potential mean-regressions, typically characterized by rapid price corrections or brief periods of sideways price action. This allows the kernel estimate to realign with actual price movements before a continuation or reversal. For those familiar with other indicators based on kernel regression, the configuration of these new features should be intuitive. The integration of these features provides an additional layer of confluence, helping users anticipate and respond to shorter-term rapid price movements. As the model is data-dependent, it will update in a real-time, non-repainting fashion as the user adjusts the settings for the envelope. The entire envelope is optional and can be toggled on or off according to user preference.

In addition to these major updates, there have been numerous minor tweaks to enhance the user interface and make the settings more generalizable to most markets by default.

٥ فبراير ٢٠٢٤

ملاحظات الأخبار

Minor update to increase the max features that can be used at once to 8.

٩ مايو ٢٠٢٤

ملاحظات الأخبار

Additional Support for Custom Feature Series
Support for 2 more custom feature series have been added, raising the total feature count to 10 and the number of custom sources to 5. This enhancement facilitates a deeper, more comprehensive ML-based analysis by leveraging additional data from custom indicators loaded within the TradingView interface. These improvements expand the model's analytical capabilities across diverse data sources for more nuanced analysis, with careful algorithmic considerations being taken to ensure performance is not compromised.

Decoupling of Mean Reversal Signals and Envelope Band Displays
In response to user feedback, the functionality for the appearance of mean reversal signals has been decoupled from the envelope band displays. This change enhances user flexibility in managing the visualization of these two distinct analytical components, allowing for the independent display of mean reversal signals without the envelope bands and vice versa. This modification enables a more personalized analytical view.

New Alerts for Mean Reversion Detection
To assist users in identifying potential market reversals, new alerts for both regular and strong upward and downward mean reversions have been introduced. These alerts facilitate strategic trading decisions by signaling when the market might be overextended, emphasizing moments for potential corrective action or sideways meandering towards the kernel.

New Daily Kernel Regression Filter
Due to popular request, a daily kernel regression filter has been added. This filter aims to smooth out noise and highlight significant trends at lower timeframes, thereby enhancing the model's focus and analytical clarity. It has been specifically designed to deactivate at the daily timeframe and above, ensuring its usage is properly restricted to lower timeframes.

New First Pullback Detection Mechanism
As a continuation of multiple collaborative efforts started with veryfid prior to his passing, an initial pass at a first pullback detection mechanism has been integrated directly into LC 2.0. By utilizing multiple boolean inputs identified by veryfid during Lorentzian Classification, this mechanism enables the identification of various conditions that may indicate first pullbacks. This development not only pays tribute to the collaborative innovation with veryfid but also enriches the model with the capability to detect subtle market movements that often are not captured by other signals.

New Alerts for First Pullback Detection
Accompanying the first pullback detection mechanism are two new alerts: one for the detection of bullish first pullbacks and another for bearish first pullbacks. With the inclusion of these alerts, the total number of new alerts introduced in this update reaches a total of 6.

١٩ أكتوبر ٢٠٢٤

ملاحظات الأخبار

Updated User Interface for Custom Sources
Custom sources have been moved to a dedicated section, making it easier to configure and manage individual indicators. The UI now offers expanded normalization options for each data set, ensuring consistency across inputs and preventing skewed backtesting results.

Enhanced Mean Reversion Detection
Performance optimizations enhance the accuracy of reversal identification, enabling more granular detection of market mean reversion signals. This update is optimized for scalping and contrarian strategies, providing potential reversal signals for short-term market movements, often before a candle closes.

New Alerts for Pullback and Mean Reversion Signals
New alert conditions for upward and downward first pullbacks, as well as mean reversion events, have been added. These alerts allow traders to respond more promptly to market shifts, improving trade execution and timeliness.

Pull Back Signal Alignment (Optional)
An optional feature aligns pullback signals more closely with price action, allowing for better synchronization between live trading and backtesting results. This ensures more accurate real-time strategy execution.

Expanded Backtest Stream
The backtest stream now includes values ranging from -5 to 5, each representing a specific signal. This expanded range allows for more nuanced backtesting and strategy development. New signals for first pullbacks and mean reversions are available for additional backtesting via a Backtest Adapter for TradingView.

Dynamically Colored Kernel Regression Ribbon
This feature highlights key moments when kernel estimates converge, diverge, or crossover, aiding in trend confirmation and market condition assessment. The visual representation with a color gradient helps traders interpret market dynamics at a glance.

Code Optimization
Optimized distance calculations, feature initialization, and signal processing boost overall efficiency, reducing processing time for faster responsiveness and smoother interaction.

Bug Fixes
Fixed issues with EMA, SMA, and Dynamic Exits, ensuring proper synchronization and consistent signals during volatile conditions. Dynamic Exits are now synchronized with market conditions, minimizing premature or delayed exits.