- ago
Hi, I am a new user at your platform. I have couple of strategies and I was wondering if there is any machine learning extensions/tools available to help me optimize my strategy. Also I am really keen to us ML to find the probability of profit of future trades based on historical patterns.
1
264
9 Replies

Reply

Bookmark

Sort
- ago
#1
It is on the way...
2
- ago
#2
It is called finantic.Learning. It will contain 37 Machine Learning (ML) algorithms for both classification and regression.
It works on top of the finantic.IndicatorSelection extension.

The workflow will be like this:
You use Indicator Profiler or finantic.IndicatorSelection to find a set of promising indicators, i.e. some indicators that have a (however small) correlation between their indicator value and future profits, i.e. show a certain predictive power.

Then you feed these (limited number of) indicators to finantic.Learning.

A machine learning algorithm will then build a model that calculates the best combination of the input indicators and predicts future profit.
2
- ago
#3
Here is a preview of the finantic.Learning extension:



It shows the combobox with algorithms to choose from and also the input page that shows the set of indicators used as inputs for the various ML algorithms.
2
- ago
#4
There is a neural net extension also:

NeuroLab:
https://www.wealth-lab.com/extension/detail/NeuroLab
1
ww58
- ago
#5
@DrKoch, can you record a videocast about finantic.Learning & IndicatorSelection extensions and what are their advantages for building profitable strategies?
2
just8
- ago
#6
A lot of great learning algorithms in that list. I can't see them all. Could you please list them all or point to the ML library being used?

Over the weekend I went through my notes and trading diary from 2009 when i was using NeuroSolutions. Best models had fewer inputs and these inputs where mostly RSI and percent change (ROC). Best ANN algorithms were time lag recurrent flavors.
Also the models did not degrade to losses as quickly as I remembered. Took 1-2 months until they were not profitable anymore. They never lost money due to good money management and position sizing. Always profitable.
Not true swing algo trading since I was babying them intraday a little to get better entries and exits.

The other important thing with ML/AI is what to predict. Neurodimension had a great idea with their optimal signal. The signals that the ANN was trained on.

https://www.youtube.com/watch?v=HeVn0riFoy4

Looks like this could be a machine learning tool and not an AI (ANN) tool.
0
- ago
#7
QUOTE:
Could you please list them all

Here is the complete list of ML algorithms supported by the upcoming finantic.Learning extension:

AdaBoost: Adaptive Boosting (Regression)
AdaBoost: Adaptive Boosting (Classification)
Averaged Perceptron (Classification)
Decission Trees (Regression)
Decission Trees (Classification)
Extremely Randomized Trees (Regression)
Extremely Randomized Trees (Classification)
Fast Forest / Random Forest (Classification)
Fast Forest / Random Forest (Regression)
Fast Tree / MART gradient boosting (Classification)
Fast Tree / MART gradient boosting (Regression)
Fast Tree with Tweedie Loss (Regression)
Generalized additive model (GAM) (Classification)
Generalized additive model (GAM) (Regression)
Gradient Boost/Absolute Loss (Regression)
Gradient Boost/Binomial Deviance (Classification)
Gradient Boost/Huber Loss (Regression)
Gradient Boost/Quantile Loss (Regression)
Gradient Boost/Square Loss (Regression)
L-BFGS Logistic Regression (Classification)
Light GBM (Classification)
Light GBM (Regression)
Linear SVM (Classification)
Local Deep SVM / LD-SVM (Classification)
Neural Net (Regression)
Neural Net (Classification)
Online Gradient Descent (OGD) (Regression)
Ordinary Least Squares (OLS) (Regression)
Random Forest (Regression)
Random Forest (Classification)
SDCA (Regression)
SDCA Binary Logistic Regression (Classification)
SDCA Non Calibrated (Classification)
Symbolic SGD Logistic Regression (Classification)

These algorithms come form two popular ML libraries: SharpLearning and Microsoft.ML.

1
- ago
#8
QUOTE:
Best models had fewer inputs and these inputs where mostly RSI and percent change (ROC)


I found a similar pattern: Do not use too many indicators as inputs. And preselect the indicatory with the best predictive power.

The finantic.IndicatorSelection extension is designed for the task to find indicators that match your trading strategy best, i.e. indicators that can predict if your trades will result in profit or loss.

Because it selects among all available indicators (several thousand variants) chances are that you fnd things that work better than RSI and ROC.
1
- ago
#9
QUOTE:
The other important thing with ML/AI is what to predict.


My suggestion is this:
Start with a trading strategy that already works reasonably well.
Then use all the trades of this strategy, i.e. the profit of each individual position.

Use the machine Learning Algorithms for classification to predict if a trade will be a winner or a looser.

Or use the machine learning algorithms for regression to predict the profit of each trade.
0

Reply

Bookmark

Sort