No one indicator is perfect. People always have their favorite indicators and maintain a bias on weighing them purely on psychological reasons other than mathematical. This technical checklist indicator collected 20 common indicators and custom ones to address the issue of a bias weighted decision.
Here, I apply machine learning using a simple sigmoid neuron network with one hidden layer and a single node to avoid artifacts. For the ease of data collection, the indicator matrix is first shown as a heatmap. Once an uptrend signal window is selected manually, an indicator matrix can be recorded in a binary format (i.e., 1 0 0 1 1 0, etc.).
For example, the following indicator matrix was retrieved from the MRNA chart (deciscion: first 5 rows, buying; last 5 rows, no buying):
This matrix is then used as an input to train the machine learning network. With a correlated buying decision matrix as an output:
<output.txt> 1 1 1 1 1 0 0 0 0 0
After training, the corrected weight matrix can be applied back to the indicator. And the display mode can be changed from a heatmap into a histogram to reveal buying signals visually.
Corresponding indicators to the weight matrix: 1. Breakout 2. Reversal 3. Crossover of ema20 and ema60 4. Crossover of ema20 and ema120 5. MACD golden cross 6. Long cycle (MACD crossover 0) 7. RSI not overbought 8. KD not overbought and crossover 9. OBV uptrend 10. Bullish gap 11. High volume 12. Breakout up fractal 13. Rebounce of down fractal 14. Convergence 15. Turbulence reversal 16. Low resistance 17. Bullish trend (blue zone) 18. Bearish trend (red zone) 19. VIX close above ema20 20. SPY close below ema20
PS. It is recommended not to use default settings but to train your weight matrix based on underlying and timeframe.
This is pretty spectacular indicator! I stumbled across it by accident but I am shocked at how good it is for swing trading. Good Job!
DEAD_HUNTER
⋅
From what I can read it sounds interesting, one thing that I am having a hard time wrapping my head around is how to use this? Does it just show a binary buy signal every once in a while based on the weights of the indicators?
DEAD_HUNTER
⋅
Also this seems like it would suffer from horrible overfitting if its retrained for each individual ticker.
purpleswan69
⋅
@DEAD_HUNTER, I think it is more appropriate to train on an individual ticker than a large group of tickers as a starting point. Group-level training implicitly assumes ergodicity which is incorrect. The only remedy for this is to treat each time series individually. There's some work showing that transfer learning would be useful for lessening overfit as you are thinking of it, but to start I think individual level analysis is most appropriate.
lockeyj
⋅
@DEAD_HUNTER, Yes. The input is a binary indicator matrix and the output is a binary buy/sell decision. After machine learning, it generates a weight matrix for these 20 indicators. They don't allow me to post the link for the python script and instruction (in Chinese). It is a bit difficult to give instruction here.
@lockeyj, hey man, any possibility of getting instructions on how to use it and train de algo?
Yas1400i
⋅
I dont know about machine learning. is this ok and ready to use it as it is by default, for BTC or EURUSD 5min?
Seungdori_
⋅
very nice! thank you for sharing this script
JC_Stoker
⋅
Cool. I have something similar that generates arrows on the chart taken from data generated by 15 or so indicators. Given me some things to think about. Thanks