Defining hyperparameters

Let's imagine that I want to optimize the period of two EMA indicators in my strategy.

This is how I would have defined them so far, by hard-coding periods as integer values; in this case 50 and 200:

@property
def slow_sma(self):
    return ta.sma(self.candles, 200)

@property
def fast_sma(self):
    return ta.sma(self.candles, 50)

To define the hyperparameters of your strategy, simply add the hyperparameters() method to your strategy. It must return a list of python dictionary objects (genes); each of which has to have these key values: name, type, min, max, and default. Here is the code for this example:

def hyperparameters(self):
    return [
        {'name': 'slow_sma_period', 'type': int, 'min': 150, 'max': 210, 'default': 200},
        {'name': 'fast_sma_period', 'type': int, 'min': 20, 'max': 100, 'default': 50},
    ]

TIP

Currently int and float are the only supported types. If you need a boolean, just use int with min: 0, max: 1.

TIP

Currently the step size of int and float can't be set. If you want a certain step size, just use int and then mutiply it with your step size inside your strategy.

slow_sma_period and fast_sma_period are the names that I chose for these two hyperparameters. It could have been anything else.

Jesse (behind the scenes) injects each hyperparameter (gene) into the self.hp property that is available in your strategy class.

Now let's rewrite my starting example to use the dynamic hyperparameters instead:

@property
def slow_sma(self):
    return ta.sma(self.candles, self.hp['slow_sma_period'])

@property
def fast_sma(self):
    return ta.sma(self.candles, self.hp['fast_sma_period'])

If I execute a backtest again, I will get the same results as before as if I was using the hard-coded code. That's because I defined 50 and 200 as default values for my hyperparameters.

# Defining hyperparameters

Defining hyperparameters