Discussing the article: "Feature Engineering With Python And MQL5 (Part IV): Candlestick Pattern Recognition With UMAP Regression"
I would like to see the great result this approach as produced
Thank you , this is a really interesting application. A couple of things that may be of help error module 'umap' has no attribute 'UMAP' you need umap-learninstalled , you can do this with a line !pip install umap-learn. if you get NameError: name 'FloatTensorType' is not defined you need to install or update onnixxmltools via !pip install onnxmltools. My data turned out very different that the data shown here , I would be interested in how everyone else gets on with the code
You're right, it is precisely dimensionality reduction, that's why I compared it to PCA 🤜🏾🔥🤛🏾
UMAP offers us more tuning parameters than traditional dimensionality reduction techniques.
linfo2 #:
Thank you , this is a really interesting application. A couple of things that may be of help error module 'umap' has no attribute 'UMAP' you need umap-learninstalled , you can do this with a line !pip install umap-learn. if you get NameError: name 'FloatTensorType' is not defined you need to install or update onnixxmltools via !pip install onnxmltools. My data turned out very different that the data shown here , I would be interested in how everyone else gets on with the code
Hey Niel.Thank you , this is a really interesting application. A couple of things that may be of help error module 'umap' has no attribute 'UMAP' you need umap-learninstalled , you can do this with a line !pip install umap-learn. if you get NameError: name 'FloatTensorType' is not defined you need to install or update onnixxmltools via !pip install onnxmltools. My data turned out very different that the data shown here , I would be interested in how everyone else gets on with the code
Thank you for sharing those solutions to commonly faced issues.
Also you've raised an important point. The same symbol can be easier to model on one broker and harder to model on a different broker.
This is partly due to the differences in data feeds that each broker has.
Some brokers collect real time ticks with minimal latency, other brokers collect price updates every minute, some only collect ticks if price levels have changed and other brokers extrapolate price levels periodically.
In brief, the difficulty of forecasting EURUSD is a function of the broker you chose.
Each broker offers a different slice, of the same market.
I want are EA for MT5 am using exness broker
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Feature Engineering With Python And MQL5 (Part IV): Candlestick Pattern Recognition With UMAP Regression.
Dimension reduction techniques are widely used to improve the performance of machine learning models. Let us discuss a relatively new technique known as Uniform Manifold Approximation and Projection (UMAP). This new technique has been developed to explicitly overcome the limitations of legacy methods that create artifacts and distortions in the data. UMAP is a powerful dimension reduction technique, and it helps us group similar candle sticks in a novel and effective way that reduces our error rates on out of sample data and improves our trading performance.
Candlestick patterns are widely used across many different trading strategies and styles by most algorithmic traders in our community. However, our understanding of these patterns is limited to the candlesticks that we have uncovered, while in truth there may be many other profitable candlestick patterns we are simply not aware of yet. Due to the wealth of information covering most modern markets, it is materially challenging for traders to be confident that they are always using the most reliable candlestick patterns available to them in their chosen market.
To alleviate this issue, we will propose a solution that potentially allows our computer to identify new candlestick patterns we were not aware of. Our proposed framework is partially comparable to a childhood game most of us should be familiar with. The game goes by different names. However, the underlying premise is the same. The game challenges the players to describe a noun using adjectives that do not contain the noun. So, for example, if the given noun were a banana, the player leading the game would give clues to his friends that best describe the banana, such as “yellow and curved”, hopefully this is intuitive for the reader to understand.
This childhood game is logically identical to the tasks we will ask our computer to perform so that we can uncover new candlestick patterns that would’ve otherwise remained hidden by the high number of dimensions our data sets tend to take these days. Analogous to the game we have just described, where the player is asked to describe a banana in 3 words or fewer, we will provide our computer with market data that has 10 columns describing the current candlestick and then ask our computer to describe the original market data in 8 columns (embeddings) or less. This is called dimension reduction.
There are many well-known dimension reduction techniques that the reader may already be acquainted with, such as Principal Components Analysis (PCA). These techniques are helpful because they guide our computer to focus on the most significant aspect of the transformed data. Today, we shall employ a technique known as Uniform Manifold Approximation And Projection (UMAP). This is a recent algorithm and, as the reader shall soon see, it can serve us by exposing non-linear relationships in our market data in a novel manner.
Author: Gamuchirai Zororo Ndawana