tensorflow machine learning standard option for algo trading but there is simpler options
Various links below show simple ways to use linear regression with R Squared. It then evolves into using SciKitLean Python Package which is relatively simple but could be effective if you are just after R Square to measure strength of the trend.
towardsdatascience.com/simple-and-multiple-linear-regression-in-python-c928425168f9
blastchart.com/Community/IndicatorGuide/Indicators/LinearRegressionRSquared.aspx
premium.working-money.com/wm/display.asp?art=340
This next link offers a look the complicated TensorFlow Model Library from Google. This is the definite standard defacto in the financial industry. I know this when I saw samples used by high end HFT shops with their workshops for the Newsweek AI conference in NYC last Nov.
Set stock data target with https://medium.com/mlreview/a-simple-deep-learning-model-for-stock-price-prediction-using-tensorflow-30505541d877 <– TensorFlow complex but offers many options
I will also say that TensorFlow offer the most choice. See below a tip below from someone in my private Telegram group. This could save you loads of time instead of wasting time by going down rabbit holes.
For crypto use, https://dashee87.github.io/deep%20learning/python/predicting-cryptocurrency-prices-with-deep-learning/
Note that this uses Keras which obviously simpler than TensorFlow. I would probably move towards TF as I get deeper into this.
Here is that useful tip:
My quick-fire tips: in general I’ve had more success constantly-retraining the ML models (every day in your case) as opposed to the usual train/test/validate method. Either that, or use ML models that are specifically designed to capture the time-varying properties of time series (these models can usually be updated in an online fashion). Never used raw prices as inputs—there is so much autocorrelation in the features that the predicted price will just end up being the last close. Instead use returns (I prefer logarithmic returns over simple returns, but it doesn’t make much difference).
Try to use a target that is directly related to the outcome of a potential trade. If that is too complicated, try having separate targets for direction and volatility. I prefer regression algos to classifiers. If possible, require that your model predicts opposite scenarios for both a long and short trade; e.g. it predicts a long trade will make money *AND* a short trade will lose money, and vice-versa. Avoid neural networks when you’re just starting out – there are too many hyperparameters to tune. Ensemble methods are fantastic, and can be very profitable
Here is my conclusion:
If you are like me just trying to figure out general trend through LSR and R-square, I tend to use these for now. As a result, I may use the SciKitLearn example for simplicity but stop there. I really don’t want to go down the various rabbit holes to experiment with more complex ML libraries. I see no reason compared to the recent sub minute study/reseach/playing around experiment at the sub minute level I started back in Dec which became a time waster with no results.
For those keeping track, if you don’t want to use Python with TensorFlow there is a simpler C++ option
https://tebesu.github.io/posts/Training-a-TensorFlow-graph-in-C++-API
This is the simplest so far but which started this whole level of research
https://stackoverflow.com/questions/893657/how-do-i-calculate-r-squared-using-python-and-numpy
You can see how simple stuff turns complex
NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!Williams % and Stochastics most reliable Matlab technical indicator
