The scope of this project is to check faisability of a fast/simple Python code for stock chart Pattern recognition with Deep Learning (neural network). From an OHLC time serie chart, the code have to be able to dectect a technical pattern in a sliding window as following:
- Focuse on just 1 technical pattern at this time (Head and Shoulders Top).
- Inputs: OHLC daily data price from a csv file and dimension of the windows (ex: 50 days) in wich the pattern have to be recognize and kind of pattern you are looking for (here: Head and Shoulders Top).
- See it as an event driven code: each day the code get a new daily data (loop on a OHLV csv file).
- After an initial calculation (1st time, we get all data needed in the windows(ex: 50 days of data)), we have to update as simple as possible the calculation each day. Calculation for each day have to be very fast (do not inverse an entire Matrix). For example, see it as a SMA (simple moving average calculation) update, you just remove the first value and add the new one (VS calculate the entire average).
- output: True or False (or 0 / 1).
- keep the code as simple as possible and as clean as possible: 1 page plus 1 page to learn and another to try.
- After learning phase pickle needed parameters in a file, and read parameters when you try.
- Learn phase: write an algorythm that create head and shoulders top charts on different windows (number of days) with different scale and noise. Use those charts in the learning phase.
- Try it and check it properly works.
- Seems to use LSTM scheme?