Optimize with Hyperopt¶

Home | Document

Installation¶

In [ ]:

!pip install hyperopt

!pip install hyperopt

Sample Strategy¶

In [1]:

# import talib.abstract as ta

from lettrade import DataFeed, Strategy, indicator as i
from lettrade.exchange.backtest import ForexBackTestAccount, let_backtest
from lettrade.indicator.vendor.qtpylib import inject_indicators

inject_indicators()


class SmaCross(Strategy):
    ema1_window = 9
    ema2_window = 21

    def indicators(self, df: DataFeed):
        # df["ema1"] = ta.EMA(df, timeperiod=self.ema1_window)
        # df["ema2"] = ta.EMA(df, timeperiod=self.ema2_window)
        df["ema1"] = df.close.ema(window=self.ema1_window)
        df["ema2"] = df.close.ema(window=self.ema2_window)

        df["signal_ema_crossover"] = i.crossover(df.ema1, df.ema2)
        df["signal_ema_crossunder"] = i.crossunder(df.ema1, df.ema2)

    def next(self, df: DataFeed):
        if len(self.orders) > 0 or len(self.positions) > 0:
            return

        if df.l.signal_ema_crossover[-1]:
            price = df.l.close[-1]
            self.buy(size=0.1, sl=price - 0.001, tp=price + 0.001)
        elif df.l.signal_ema_crossunder[-1]:
            price = df.l.close[-1]
            self.sell(size=0.1, sl=price + 0.001, tp=price - 0.001)


lt = let_backtest(
    strategy=SmaCross,
    datas="example/data/data/EURUSD_5m-0_10000.csv",
    account=ForexBackTestAccount,
    # plotter=None,
)

# import talib.abstract as ta from lettrade import DataFeed, Strategy, indicator as i from lettrade.exchange.backtest import ForexBackTestAccount, let_backtest from lettrade.indicator.vendor.qtpylib import inject_indicators inject_indicators() class SmaCross(Strategy): ema1_window = 9 ema2_window = 21 def indicators(self, df: DataFeed): # df["ema1"] = ta.EMA(df, timeperiod=self.ema1_window) # df["ema2"] = ta.EMA(df, timeperiod=self.ema2_window) df["ema1"] = df.close.ema(window=self.ema1_window) df["ema2"] = df.close.ema(window=self.ema2_window) df["signal_ema_crossover"] = i.crossover(df.ema1, df.ema2) df["signal_ema_crossunder"] = i.crossunder(df.ema1, df.ema2) def next(self, df: DataFeed): if len(self.orders) > 0 or len(self.positions) > 0: return if df.l.signal_ema_crossover[-1]: price = df.l.close[-1] self.buy(size=0.1, sl=price - 0.001, tp=price + 0.001) elif df.l.signal_ema_crossunder[-1]: price = df.l.close[-1] self.sell(size=0.1, sl=price + 0.001, tp=price - 0.001) lt = let_backtest( strategy=SmaCross, datas="example/data/data/EURUSD_5m-0_10000.csv", account=ForexBackTestAccount, # plotter=None, )

Optimize¶

In [2]:

# define a search space
from hyperopt import fmin, tpe, space_eval, Trials
from hyperopt import hp


lettrade_model = lt.optimize_model()


def train_model(params):
    # Model
    result = lettrade_model(params)

    # Score
    return -result["equity"]


# Hyperopt
search_space = {
    "ema1_window": hp.uniformint("ema1_window", 5, 25, q=1),
    "ema2_window": hp.uniformint("ema2_window", 5, 50, q=1),
}

trials = Trials()
best_params = fmin(
    train_model,
    search_space,
    algo=tpe.suggest,
    max_evals=1_000,
    trials=trials,
)

# define a search space from hyperopt import fmin, tpe, space_eval, Trials from hyperopt import hp lettrade_model = lt.optimize_model() def train_model(params): # Model result = lettrade_model(params) # Score return -result["equity"] # Hyperopt search_space = { "ema1_window": hp.uniformint("ema1_window", 5, 25, q=1), "ema2_window": hp.uniformint("ema2_window", 5, 50, q=1), } trials = Trials() best_params = fmin( train_model, search_space, algo=tpe.suggest, max_evals=1_000, trials=trials, )

100%|██████████| 1000/1000 [01:17<00:00, 12.97trial/s, best loss: -10187.82]

In [3]:

hyperparams = space_eval(search_space, best_params)

print(best_params)
print(hyperparams)

hyperparams = space_eval(search_space, best_params) print(best_params) print(hyperparams)

{'ema1_window': 17.0, 'ema2_window': 8.0}
{'ema1_window': 17, 'ema2_window': 8}

Plot¶

In [4]:

lt.plotter.heatmap(x="ema1_window", y="ema2_window", z="equity")

lt.plotter.heatmap(x="ema1_window", y="ema2_window", z="equity")

In [5]:

lt.plotter.contour(x="ema1_window", y="ema2_window", z="equity")

lt.plotter.contour(x="ema1_window", y="ema2_window", z="equity")

Plot plotly¶

Tutorial

Init Plotly environment¶

In [6]:

import plotly.io as pio

pio.renderers.default = "notebook"
pio.templates.default = "plotly_dark"

import plotly.io as pio pio.renderers.default = "notebook" pio.templates.default = "plotly_dark"

In [7]:

import pandas as pd
import numpy as np


def unpack(x):
    if x:
        return x[0]
    return np.nan


# We'll first turn each trial into a series and then stack those series together as a dataframe.
df = pd.DataFrame([pd.Series(t["misc"]["vals"]).apply(unpack) for t in trials])
# Then we'll add other relevant bits of information to the correct rows and perform a couple of
# mappings for convenience
df["loss"] = [t["result"]["loss"] for t in trials]
df["trial_number"] = df.index


df["win"] = -df["loss"]
df

import pandas as pd import numpy as np def unpack(x): if x: return x[0] return np.nan # We'll first turn each trial into a series and then stack those series together as a dataframe. df = pd.DataFrame([pd.Series(t["misc"]["vals"]).apply(unpack) for t in trials]) # Then we'll add other relevant bits of information to the correct rows and perform a couple of # mappings for convenience df["loss"] = [t["result"]["loss"] for t in trials] df["trial_number"] = df.index df["win"] = -df["loss"] df

Out[7]:

	ema1_window	ema2_window	loss	trial_number	win
0	19.0	10.0	-10000.68	0	10000.68
1	10.0	7.0	-10062.72	1	10062.72
2	18.0	40.0	-9899.16	2	9899.16
3	9.0	8.0	-10052.78	3	10052.78
4	12.0	27.0	-9954.98	4	9954.98
...	...	...	...	...	...
995	15.0	15.0	-10000.00	995	10000.00
996	16.0	5.0	-10029.46	996	10029.46
997	18.0	13.0	-10020.16	997	10020.16
998	20.0	9.0	-9970.36	998	9970.36
999	13.0	8.0	-10140.46	999	10140.46

1000 rows × 5 columns

Type 1¶

In [8]:

from plotly import express as px

fig = px.scatter(df, x="trial_number", y="win")
fig.show()

from plotly import express as px fig = px.scatter(df, x="trial_number", y="win") fig.show()

Type 2¶

In [9]:

import plotly.express as px


fig = px.density_contour(
    df,
    x="ema1_window",
    y="ema2_window",
    z="win",
    histfunc="max",
)
fig.update_traces(contours_coloring="fill", contours_showlabels=True)
fig.show()

import plotly.express as px fig = px.density_contour( df, x="ema1_window", y="ema2_window", z="win", histfunc="max", ) fig.update_traces(contours_coloring="fill", contours_showlabels=True) fig.show()

Type 3¶

In [10]:

import plotly.express as px

fig = px.density_heatmap(
    df,
    x="ema1_window",
    y="ema2_window",
    z="win",
    nbinsx=20,
    nbinsy=40,
    histfunc="max",
    color_continuous_scale="Viridis",
)
fig.show()

import plotly.express as px fig = px.density_heatmap( df, x="ema1_window", y="ema2_window", z="win", nbinsx=20, nbinsy=40, histfunc="max", color_continuous_scale="Viridis", ) fig.show()