Optimizer creation

Suppose we want to create a custom optimizer for a StocksSet. How can we do it? In this tutorial the structure of the optimizers will be shown, naming every definition needed for an optimizer to work. The tutorial then will finish with a creation of a dummy optimizer.

Lets see then how does an optimizer works.

The first thing to consider its the base class of every optimizer. There is a class named OptimizerABC that has the following method:

def optimize(self, ss):
    weights, metadata = self._calculate_weights(ss)
    return ss.copy(weights=weights, optimizer=metadata)

Given this, we determine the need to implement a _calculate_weights function that provides the necessary data to create a new StocksSet instance with updated weights. That method is the core method of every optimizer. The weights definition.

Say for example that we want to create a new optimizer that only returns a normalized vector of weights. We can do that by making something like this.

from garpar.optimize.opt_base import OptimizerABC
import numpy as np

class MyOptimizer(OptimizerABC):
    family = "MyFamily"

    def _calculate_weights(self, ss):
        return np.ones(len(ss.weights))/len(ss.weights), {"name": "my_model"}

Note that we have defined an attribute named family, which represents the set of models it belongs to. For example, in this context, mean-variance is a family.

Now that we have defined an optimizer, let’s examine its output for a StocksSet instance.

from garpar.datasets import make_risso_normal

opt = MyOptimizer()
ss = make_risso_normal(random_state=42, stocks=5, days=10)
opt.optimize(ss)

Stocks	S0[W 0.2, H 0.5]	S1[W 0.2, H 0.5]	S2[W 0.2, H 0.5]	S3[W 0.2, H 0.5]	S4[W 0.2, H 0.5]
Days
0	100.000000	100.000000	100.000000	100.000000	100.000000
1	98.805864	100.935769	99.326480	100.900464	100.649803
2	99.723073	99.709073	98.300671	99.884634	101.809271
3	100.481068	98.845525	99.570943	98.878556	100.721910
4	99.870674	97.846148	100.461281	99.783308	99.528206
5	101.124337	96.897283	101.310699	98.637364	100.581171
6	102.006936	95.770467	100.074504	99.746865	101.208367
7	103.332098	94.844957	101.280570	100.450735	100.356336
8	104.518001	94.306958	102.217987	99.473017	101.303750
9	105.458859	93.176509	100.824345	100.715559	100.136298
10	106.641517	92.403867	101.862962	99.497536	99.235579

11 days x 5 stocks - W.Size 5

As expected, this optimizer assings weights with equal values.

Thank you for visiting the tutorials for Garpar! We recommend to look in the API section for any documentation you might need.