SEE ALSO Machine Learning Deep Detection

Objetive:

Obtain N trading strategies with technical patterns with Random Forest python-sklearn, one strategy per action, be able to generate N strategies for each action. Only candles with 1 day format

For example, this strategy has been generated by ML, with stock_strategy_random_forest_sklearn @Leci37 also generates the Pine Script™ v5 code for use in TraderView https://es.tradingview.com/chart/?symbol=NYSE%3AALL (stock ex: Allstate Corporation) here. ALL_d5_q0.22_id1204_.pine

For models with deep learning and more than 100 technical patterns are recommended TensorFlow deep learning

These strategies should also have:

Real time alert:

• Know point of buy
• StopLoss Location and StopLoss Updates
• Report closing and sale To see the alerts, some of them are shown in the account https://twitter.com/Whale__Hunters .

At the code level:

• Generate Pine Script™ v5 code for use in TraderView, example AAPL_d5_q0.217_TraderView.pine
• Easy implementation with auto trader API like Alpaca or deGiro (ask about it)
• Record of SL operations and updates. Example AAPL d_result/stra_simulator/AAPL.csv
• Evaluate the quality of the strategy d_result/history/AAPL.csv with python
• Calculate the associated technical patterns (only the konkorde pattern get_konkorde_params_GOOD( will be used in this version)
• Ground True (Y_target the variable to predict), in this case where to buy and where to sell, simple, the candlesticks where there are gains greater than 1% (defined in the variable PER_VALEU_CHANGE = 1). You can see it in the method get_GT_day_candle(df_kon) and extra rolling_buy_sell_val_BUY(df_ind)

By default it works with these two stocks stocks_list = ["AAPL", "SHIP"]

Important: The RamdonForest Model will return a number between 0 and 2, with 2 being most secure buy and 0 being sell. Telegram GROUP: https://t.me/+3oG6U_hp93I2M2Ix

Execution:

Install requirements

S_00_dowload_data_API_yahoo.py Download the historical values of OLHLV with the Yahoo API (you can use others such as alpaca but since you work with daily values it is not necessary)

S_01_train_and_tunner_.py Collect OLHLV data Calculate desired technical patterns get_konkorde_params(). Get the ground true GT to predict get_GT_day_candle(). Split the data train_test_split To fine-tune the input parameters to RamdonForest, you have to train it with multiple values with dict_paramGrid_rf The results of all models are saved in d_price/RF/AAPL_248.rfmodel Some PineScrip is generated, to see it in traderview with the best models d_price/pine_tree/AAPL_d5_q0.218_id2163_.pine

S_01b_train_deep_tunner_OPtional.py the same as the previous point with more combinations in dict_paramGrid_rf

S_02_RF_model_eval_and_save.py takes the 20 best combinations from point S_01 and reevaluates and trains them. RF models are saved in d_price/RF/AAPL_.rfmodel A file is generated that brings together all the evaluations of all the actions d_price/RF/aa_RF_full_eval.csv, check the column final_score

S_03_strategy_simulator_past.py The RF model gives us the buy points, but how do we adjust the SL when the operation should be completed? To answer these questions is the Strategy_TW_class.Strategy.update() code. To collect the complete simulation history of the generated strategy, the files d_result/stra_simulator are used with the history of all the operations that this code would have performed. They can also be seen in trader view with the files d_result/pine_TW/AAPL_d5_q0.218_id2163_pine.pine

S_04_rf_eval_Realtime.py evaluate in real time and in the future, the operations in today's are recorded in d_result/win_loss_today_2024_03_11.csv. It is implementable with automatic traders API (for alpaca deGiro, etc. consult the author) there are certain alerts in https://twitter.com/Whale__Hunters .

Example pine script

Will return a number between 0 and 2, with 2 being most safe to buy and 0 to be most safe to sell:

//@version=5
strategy("konk_AAPL_q0.199", overlay=true, margin_long=100, margin_short=100, pyramiding=5)
decision_tree_0(azul, marron, verde, media, azul_mean, verde_mean, marron_mean, verde_azul, verde_media, media_azul) =>
	var float ret = -1 // # DecisionTreeRegressor(criterion='poisson', max_depth=5, max_features=0.7,
	if( marron_mean <= 103.606 )
		if( azul_mean <= -0.193724 )
			if( verde <= -12.581 )
				if( marron_mean <= 37.7765 )
					if( azul <= 1.83462 )
						ret := 1.555556
					if( azul > 1.83462 )
						ret := 0.600000
				if( marron_mean > 37.7765 )
					ret := 0.250000
			if( verde > -12.581 )
				if( verde <= 33.8535 )
					if( marron <= 34.2398 )
						ret := 0.711538
					if( marron > 34.2398 )
						ret := 0.181818
				if( verde > 33.8535 )
					if( media_azul <= 35.982 )
						ret := 1.529412
					if( media_azul > 35.982 )
						ret := 0.921569
		if( azul_mean > -0.193724 )
			if( azul <= 40.1892 )
				if( media <= 52.4629 )
					if( marron_mean <= 44.048 )
						ret := 1.193396
					if( marron_mean > 44.048 )
						ret := 1.555556
				if( media > 52.4629 )
					if( media <= 63.4882 )
						ret := 0.827160
					if( media > 63.4882 )
						ret := 1.125551
			if( azul > 40.1892 )
				if( media_azul <= -23.329 )
					ret := 1.714286
				if( media_azul > -23.329 )
					if( media <= 60.413 )
						ret := 0.238095
					if( media > 60.413 )
						ret := 1.166667
	if( marron_mean > 103.606 )
		if( media_azul <= 65.3109 )
			ret := 0.666667
		if( media_azul > 65.3109 )
			if( verde_azul <= 110.687 )
				if( media <= 102.936 )
					if( verde_media <= 27.9566 )
						ret := 1.750000
					if( verde_media > 27.9566 )
						ret := 1.166667
				if( media > 102.936 )
					if( verde_mean <= 117.225 )
						ret := 1.000000
					if( verde_mean > 117.225 )
						ret := 1.473684
			if( verde_azul > 110.687 )
				if( verde_azul <= 115.788 )
					if( marron <= 111.401 )
						ret := 1.133333
					if( marron > 111.401 )
						ret := 0.142857
				if( verde_azul > 115.788 )
					if( marron <= 105.636 )
						ret := 1.866667
					if( marron > 105.636 )
						ret := 1.236220
	
    ret //return

Randon Forest visualization, final sheet is labeled: buy, none, sell

Why has Randon Forest made a decision?

You can view plot the model in a graph by uncommenting:

tree_to_code(rf_model.estimators_[0], rf_model.feature_names_in_)
r = export_text(rf_model.estimators_[0], feature_names=list(rf_model.feature_names_in_))
exported_text, sas_text, py_text, code_TVW= export_code(eval_model, 0, list(rf_model.feature_names_in_))

Notes:

• Be careful with the risk of overfitting, 4 or 5 branches at most are used "max_depth": [4,5] , the smallest possible are recommended.
• For more technical indicators in python you can find 200 here, look the funtion extract_features(df: pd.DataFrame,extra_columns =False, shift=150, debug=False) inside https://github.com/Leci37/TensorFlow-stocks-prediction-Machine-learning-RealTime/blob/master/features_W3_old/v3.py some of the technical indicators take future data, be careful.
• If you want to obtain the RF code in sas, TraderView and python see the function exported_text, sas_text, py_text, code_TVW = export_code(rf_mod.estimators_[0], 0, list(rf_mod.feature_names_in_))