Lazy Trading Part 6: Detect Market status with AI

Market types – the first key to building a winning Forex trading system Expecting the same system to work in all market types is the definition of insanity.” – Van K. Tharp

We will use the concept shown in the book to manually identify market periods. According to this book there are 6 main market types:

1. Bull normal
2. Bull volatile
3. Bear normal
4. Bear volatile
5. Sideways quiet
6. Sideways volatile

Theory on how to detect market types with Deep Learning

Building understanding on the h2o infrasctucture

Getting the code

The 2(3) ways to get the data:

1. Use dataset prepared with interactive visualization

2. Directly use prepared dataset

3. Collect the data while the model is in production

Lecture will demonstrate how to get financial data log from MQL to the Computer File System:

• Data Writer v 4.02

• Use function writeDataMacd()

• call function every bar

• Collect data for H1 timeframe

Q. Which indicator is best? Which indicator should I use? I want to use this system on several timeframes, will I succeed?

This lecture is intended to answer few frequently asked questions

Content:

1. This is not a trading advise

2. Any indicator that...

3. Experiment but be prudent

This is not a trading advise

Authors of this course once again recommend to use provided content as a way to learn computer and data science. In fact classification models could be used for pattern classification in a number of applications like:

• classify product physical property by using a sound waves data

• classify temperature profile from machinery processes

• etc

Any indicator that...

Use of macd indicator shown in the course should be used with caution. The absolute value of the indicator is dependent of the asset price. This may influence performance of the model. Authors of the course may recommend to use other indicators that do not have such dependency. For example indicators that have min max range that:

• is from 0 to 100

• or from -100 to 100

• or from -1 to +1

• etc

Experiment but be prudent

Authors of the course would definitely appreciate whenever students will use material of this course to experiment and re-create proposed software. Experimenting is known to be the most effective way to master the subject. However handling complex code may be complicated. Please be mindful to first test the concept on something simple and only after decide whether it is a good option to expand...

Thank you for understanding

2020-11-22

Collecting data in the interactive way:

1. Use provided data sample

2. Modular code to extract and aggregate the data

Visualize data as 3D. Using R library(plotly)

Visualize data matrix in 3D graph

We talk about 2 ways of getting the dataset. Particular attention is given on how to:

• convert one column to factor

• inspect number of factor using function summary(df$column)

Learning Goals

• Train Classification Deep Learning Model

  • Use classified data

  • Use prepared function

• How function works

  • Random neural network structures

  • Model performance

  • Save the model

Use script to build deep learning model and write it to the disk

Reminder to update / retrain Deep Learning models when updating version of h2o package

What is inside the function:

• random neural network structures

• checking model metrics

• logging results of modeling internally

• ...

Propose a way to automate this task to build the model, why?

• we may update h2o package

• to try to improve the model by using automatically collected new data

Deploy Script aim is to:

• classify current market type for 28 assets based on the values of the latest indicator

• record new data for high selected patterns where confidence of classification is high

• Look into internals of the function mt_evaluate

• Where to find documentation, examples

This lecture shows what needs to be modified in order to run this code locally and reproduce provided example

Run the script

03_Score_Data.R

Checking how automatic pattern recognition is effective to recognize market type:

• watch files in the sandboxes

• watch the charts

• compare market type visually also looking on 'confidence' values inside files

adapt this script and automate to run this program in Windows task scheduler

• run on Sunday before market is opened

• run weekly on Monday - Tue- Wed - Thu - Fr

• Repeat every 1 hour

• start running at 0:10

Look at the function mt_evaluate

This may be useful for any data science project:

• collect predictor data while model is used

• do that on specified condition

• store this info persistently (e.g. write to the file)

Reminder that Functions in R package could be understood by reproducing an example

Explaining the reasons of creating this chapter:

• creating a User Interface for quick data visualization and check

• balancing classes option

• suggestion to create an algorithm

Shiny App Prototype for data review

Review how Shiny App works to manually inspect and save data 'aside'

how to add new data for model update

example of creating algorithm blueprint

Read Market Type information in MQL4

Focus of this chapter is to understand the mechanism of using Market Type information in our Trading Robots:

• How to read Market Type information in MQL4 program?

• How to log Market Type information at the moment of opening new Trades (log to use Reinforcement Learning and define best Market Type for a Trading Robot)

Note that this course will only focus on generating the 'output', the 'demo trading experiment' will be made in the course 7

Reading the data out of the file in the sandbox

Reading 'double' values from flat files to MetaEditor platforms

Code shown inside video lectures

How Market Type info can be used inside Trading Robot

Describe the problem and how we will be solving that!

Show Jumping Monkey Simulation to illustrate the problem

Simulating self-organizing system using R software

Reviewing the code capable to log the Market Status data of the system to the file

Generate data for the Reinforcement Learning Problem. Adding Market Status information to the Trading Results

In this lecture we will be performing Reinforcement Learning on the obtained dataset. Idea will be to create a model. Model will define the policy suggesting optimal Market Type for the Trading System

A note about additional code to find best control parameters

In this lecture we will see how to use Reinforcement Learning policy inside of the Trading Terminal

Summary of steps taken in this section

This lecture outlays learned programming concepts with particular focus on Data Science Modelling

This lecture provides a bonus coupon to the next course for lazy traders