SplineHyperOptLoss, for easier hyperopt

2021-09-05 22:34:47 +02:00 · 2021-09-05 22:34:47 +02:00 · 548d83fe19
commit 548d83fe19
parent 01b5fe9f97
1 changed files with 129 additions and 0 deletions
--- a/freqtrade/optimize/hyperopt_loss_splinehyperoptloss.py
+++ b/freqtrade/optimize/hyperopt_loss_splinehyperoptloss.py
@ -0,0 +1,129 @@
+import logging
+from datetime import datetime
+
+from freqtrade.data.btanalysis import calculate_max_drawdown
+from freqtrade.optimize.hyperopt import IHyperOptLoss
+from pandas import DataFrame
+from scipy import interpolate
+
+logger = logging.getLogger(__name__)
+
+import numpy as np
+
+# HyperOpt Loss function, it is difficult to put the various optimizations into one number.
+# To make it easier, you can specify one table per optimization here. Missing values are interpolated.
+# In this way it is possible to create your own loss curves for each value quite easily.
+
+#
+# this function also removes extrem vales from the results, to optimize on the major mass to avoid
+# single trades/pairs to dominate
+#
+
+class SplineHyperOptLoss(IHyperOptLoss):
+
+    @staticmethod
+    def hyperopt_loss_function(results: DataFrame, trade_count: int,
+                               min_date: datetime, max_date: datetime,
+                               *args, **kwargs) -> float:
+
+        def remove_outliners(x) -> np.array:
+            x = x[x.between(x.quantile(.15), x.quantile(.85))]
+            return x
+
+        def execute_table(table: dict, x: float) -> float:
+            x_points = [float(k) for k in table.keys()]
+            y_points = list(table.values())
+            tck = interpolate.splrep(x_points, y_points, k=2)
+            return float(interpolate.splev(x, tck))
+
+        try:
+            max_drawdown = calculate_max_drawdown(results)[0]
+        except:
+            max_drawdown = 0.0
+
+        #  profit_sum = results['profit_ratio'].sum()
+        profit = remove_outliners(results['profit_ratio']).sum()
+        profit_mean = remove_outliners(results['profit_ratio']).mean()
+        profit_median = remove_outliners(results['profit_ratio']).median()
+        trade_duration = remove_outliners(results['trade_duration']).mean()
+        trade_count = len(results)
+
+        ret = 0
+        profit_table = {
+            '-1.0': -1000,
+            '0.2': 50,
+            '0.3': 100,
+            '1.0': 500,
+            '2.0': 1000,
+            '100.0': 10000,
+        }
+        r = execute_table(profit_table, profit)
+        logger.debug(f'profit: {profit}')
+        logger.debug(f'profit ret: {r}')
+        ret += r
+
+        profit_mean_table = {
+            '-1.0': -1000,
+            '0.01': 50,
+            '0.02': 500,
+            '0.03': 1000,
+            '0.04': 500,
+            '0.05': 50,
+        }
+        r = execute_table(profit_mean_table, profit_mean)
+        logger.debug(f'profit_mean: {profit_mean}')
+        logger.debug(f'profit_mean ret: {r}')
+        ret += r
+
+        profit_median_table = {
+            '-1.0': -1000,
+            '0.01': 50,
+            '0.02': 1000,
+            '0.03': 1000,
+            '0.04': 50,
+            '0.05': 10,
+        }
+        r = execute_table(profit_median_table, profit_median)
+        logger.debug(f'profit_median: {profit_median}')
+        logger.debug(f'profit_median ret: {r}')
+        ret += r
+
+        trade_duration_table = {
+            str(1): -10,
+            str(5): 5,
+            str(60): 500,
+            str(60 * 5): 250,
+            str(60 * 10): 150,
+            str(60 * 20): 50,
+            str(60 * 2000): -10,
+        }
+        r = execute_table(trade_duration_table, trade_duration)
+        logger.debug(f'trade_duration: {trade_duration}')
+        logger.debug(f'trade_duration ret: {r}')
+        ret += r
+
+        trade_count_table = {
+            '0': 0,
+            '20': 100,
+            '100': 500,
+            '300': 1000,
+            '500': 500,
+            '1000': 50,
+        }
+        r = execute_table(trade_count_table, trade_count)
+        logger.debug(f'trade_count: {trade_count}')
+        logger.debug(f'trade_count ret: {r}')
+        ret += r
+
+        max_drawdown_table = {
+            '0': 1000,
+            '0.1': 3000,
+            '0.4': 500,
+            '0.8': -5000,
+        }
+        r = execute_table(max_drawdown_table, max_drawdown)
+        logger.debug(f'max_drawdown: {max_drawdown}')
+        logger.debug(f'max_drawdown ret: {r}')
+        ret += r
+
+        return -ret