I am working on a trading bot based on the famously known Talib technical analisys library and got sidetracked with learning about candlestick pattern recognition and even if there is plenty of information on the internets I just wanted to make an introductory piece on patterns and how I am planning on building an AI bot that will treat the market as a DNA chain with repeating patterns that will allow us to predict the next candlestick based on past probability and some machine learning. Like REHHD.GBLHM.GBHHU.GBHHU would be the sequence for "Three White Soldiers" after a red "Doji" then using years of historic data for training the oracle would allow us to predict the market (probabilistically).
For that we will start with the basic candlestick patterns individually. If we classify them by trend, body size, shadow and position we may form a sequencing pattern that may be easier for bots to recognize. Next we will classify them in groups of two and three building our database of knowledge by probability of occurrence.
If anybody is interested in Machine Learning and pattern recognition just join the discussion.
* One year of training would give us over 500k 1min ticks, two years for one million ticks.
