Source code for so4gp.algorithms.graank_ga

# -*- coding: utf-8 -*-
# SPDX-License-Identifier: GNU GPL v3
# This file is licensed under the terms of the GNU GPL v3.0.
# See the LICENSE file at the root of this
# repository for complete details.


import json
import time
import numpy as np
from ..data_gp import DataGP
from .numeric_ss import NumericSS


[docs] class GeneticGRAANK(DataGP):
[docs] def __init__(self, *args, max_iter=1, n_pop=5, pc=0.5, gamma=1.0, mu=0.9, sigma=0.9, **kwargs): """ Extract gradual patterns (GPs) from a numeric data source using the Genetic Algorithm approach (proposed in a published paper by Dickson Owuor). A GP is a set of gradual items (GI), and its quality is measured by its computed support value. For example, given a data set with 3 columns (age, salary, cars) and 10 objects. A GP may take the form: {age+, salary-} with a support of 0.8. This implies that 8 out of 10 objects have the values of column age 'increasing' and column 'salary' decreasing. In this approach, we assume that every GP candidate may be represented as a binary gene (or individual) that has a unique position and cost. The cost is derived from the computed support of that candidate, the higher the support value, the lower the cost. The aim of the algorithm is to search through a population of individuals (or candidates) and find those with the lowest cost as efficiently as possible. :param args: [required] a data source path of Pandas DataFrame, [optional] minimum-support, [optional] eq :param max_iter: [optional] maximum_iteration, default is 1 :type max_iter: int :param n_pop: [optional] initial individual population, default is 5 :type n_pop: int :param pc: [optional] children proportion, default is 0.5 :type pc: float :param gamma: [optional] cross-over gamma ratio, the default is 1 :type gamma: float :param mu: [optional] mutation mu ratio, default is 0.9 :type mu: float :param sigma: [optional] mutation sigma ratio, default is 0.9 :type sigma: float >>> from so4gp.algorithms as GeneticGRAANK >>> import pandas >>> >>> dummy_data = [[30, 3, 1, 10], [35, 2, 2, 8], [40, 4, 2, 7], [50, 1, 1, 6], [52, 7, 1, 2]] >>> dummy_df = pandas.DataFrame(dummy_data, columns=['Age', 'Salary', 'Cars', 'Expenses']) >>> >>> mine_obj = GeneticGRAANK(dummy_df, 0.5, max_iter=3, n_pop=10) >>> result_json = mine_obj.discover() >>> # print(result['Patterns']) >>> print(result_json) # doctest: +SKIP {"Algorithm": "GA-GRAANK", "Best Patterns": [[["Age+", "Salary+", "Expenses-"], 0.6]], "Invalid Count": 12, "Iterations": 2} """ super(GeneticGRAANK, self).__init__(*args, **kwargs) self._max_iteration: int = max_iter self._parent_pop: int = n_pop self._children_pop: float = pc self._gamma: float = gamma self._mu: float = mu self._sigma: float = sigma
def _crossover(self, p1: NumericSS.Candidate, p2: NumericSS.Candidate) -> tuple[NumericSS.Candidate, NumericSS.Candidate]: """ Crosses over the genes of 2 parents (an individual with a specific position and cost) to generate 2 different offsprings. :param p1: The parent-1 individual :param p2: The parent-2 individual :return: Two offsprings (children) """ c1 = NumericSS.Candidate() c2 = NumericSS.Candidate() alpha = np.random.uniform(0, self._gamma, 1) c1.position = float(alpha * p1.position + (1 - alpha) * p2.position) c2.position = float(alpha * p2.position + (1 - alpha) * p1.position) return c1, c2 def _mutate(self, x: NumericSS.Candidate): """ Mutates an individual's position to create a new and different individual. :param x: The existing individual :return: A new individual """ y = NumericSS.Candidate(position=x.position, cost=x.cost) str_x = str(int(y.position) if y.position is not None else 0) flag = np.random.rand(*(len(str_x),)) <= self._mu ind = np.argwhere(flag) str_y = "0" for i in ind: val = float(str_x[i[0]]) val += self._sigma * np.random.uniform(0, 1, 1) if i[0] == 0: str_y = "".join(("", "{}".format(int(val)), str_x[1:])) else: str_y = "".join((str_x[:i[0] - 1], "{}".format(int(val)), str_x[i[0]:])) str_x = str_y y.position = int(str_y) return y def discover(self): """ Uses genetic algorithm to find GP candidates. The candidates are validated if their computed support is greater than or equal to the minimum support threshold specified by the user. :return: JSON object """ # Prepare data set start = time.time() self.fit_bitmap() self.clear_gradual_patterns() if self.valid_bins is None: return [] # Initialize search space s_space = NumericSS.initialize_search_space(self.valid_bins, self._parent_pop, self._max_iteration) if s_space is None: return [] num_children = int(np.round(self._children_pop * self._parent_pop / 2) * 2) # Number of children np.round is used to get an even number repeated = 0 while s_space.counter < self._max_iteration: c_pop = [] # Children population for _ in range(num_children // 2): # Select Parents q = np.random.permutation(self._parent_pop) p1 = s_space.pop[q[0]] p2 = s_space.pop[q[1]] # a. Perform Crossover c1, c2 = self._crossover(p1, p2) NumericSS.evaluate_candidate(c1, s_space, self.valid_bins) NumericSS.evaluate_candidate(c2, s_space, self.valid_bins) # b. Perform Mutation c1 = self._mutate(c1) c2 = self._mutate(c2) NumericSS.evaluate_candidate(c1, s_space, self.valid_bins) NumericSS.evaluate_candidate(c2, s_space, self.valid_bins) # c. Add Offsprings to c_pop c_pop.append(c1) c_pop.append(c2) # Merge, Sort and Select s_space.pop += c_pop s_space.pop = sorted(s_space.pop, key=lambda x: x.cost) s_space.pop = s_space.pop[0:self._parent_pop] # Evaluate GP _, repeated = NumericSS.evaluate_gradual_pattern(repeated, s_space, self) for gp in s_space.best_patterns: self.add_gradual_pattern(gp) duration = time.time() - start out_dict: dict[str, str | list] = { "Algorithm": "GA-GRAANK", # "Memory Usage (MiB)": f{mem_use)}" "Initial Population": f"{self._parent_pop}", "Children Proportion": f"{self._children_pop}", "Crossover Gamma": f"{self._gamma}", "Mutation Mu": f"{self._mu}", "Mutation Sigma": f"{self._sigma}", "Number of iterations": s_space.iter_count, "Run-time": f"{duration:.6f} seconds"} self.generate_output_files(out_dict) out_dict.update({"Best Patterns": s_space.str_best_gps, "Invalid Count": str(s_space.invalid_count)}) out: object = json.dumps(out_dict, indent=4) return out