Add summary

docs/Summary.org

@@ -0,0 +1,143 @@
+#+TITLE: Práctica 2
+#+SUBTITLE: Metaheurísticas
+#+AUTHOR: Amin Kasrou Aouam
+#+DATE: 2021-06-22
+#+PANDOC_OPTIONS: template:~/.pandoc/templates/eisvogel.latex
+#+PANDOC_OPTIONS: listings:t
+#+PANDOC_OPTIONS: toc:t
+#+PANDOC_METADATA: lang=es
+#+PANDOC_METADATA: titlepage:t
+#+PANDOC_METADATA: listings-no-page-break:t
+#+PANDOC_METADATA: toc-own-page:t
+#+PANDOC_METADATA: table-use-row-colors:t
+#+PANDOC_METADATA: colorlinks:t
+#+PANDOC_METADATA: logo:/home/coolneng/Photos/Logos/UGR.png
+#+LaTeX_HEADER: \usepackage[ruled, lined, linesnumbered, commentsnumbered, longend]{algorithm2e}
+* Práctica 2
+
+** Introducción
+
+En esta práctica, usaremos distintos algoritmos de búsqueda, basados en poblaciones, para resolver el problema de la máxima diversidad (MDP). Implementaremos:
+
+- Algoritmo genético
+- Algoritmo memético
+
+** Algoritmos
+
+*** Genético
+
+Los algoritmos genéticos se inspiran en la evolución natural y la genética. Generan un conjunto de soluciones inicial (i.e. población), seleccionan un subconjunto de individuos sobre los cuales se opera, hacen operaciones de recombinación y mutación, y finalmente reemplazan la población anterior por una nueva.
+
+El procedimiento general del algoritmo queda ilustrado a continuación:
+
+\begin{algorithm}
+    \KwIn{A list $[a_i]$, $i=1, 2, \cdots, n$, that contains the population of individuals}
+    \KwOut{Processed list}
+
+    $P(t) \leftarrow initializePopulation()$
+    $P(t) \leftarrow evaluatePopulation()$
+
+    \While{$\neg stop  condition $}{
+        $t = t + 1$
+
+        $parents \leftarrow selectParents(P(t-1))$
+
+        $offspring \leftarrow recombine(parents)$
+
+        $offspring \leftarrow mutate(offspring)$
+
+        $P(t) \leftarrow replacePopulation(P(t-1), offspring)$
+
+        $P(t) \leftarrow evaluatePopulation()$
+    }
+    \KwRet{$P(t)$}
+\end{algorithm}
+
+Procedemos a la implementación de 4 variantes distintas, según 2 criterios:
+
+**** Criterio de reemplazamiento
+- *Generacional*: la nueva población reemplaza totalmente a la población anterior
+- *Estacionario*: los dos mejores hijos reemplazan los dos peores individuos en la población anterior
+**** Operador de cruce
+- *Uniforme*: mantiene las posiciones comunes de ambos padres, las demás se eligen de forma aleatoria de cada padre (requiere reparador)
+- *Posición*: mantiene las posiciones comunes de ambos padres, elige el resto de elementos de cada padre y los baraja. Genera 2 hijos.
+
+*** Memético
+
+Los algoritmos meméticos surgen de la hibridación de un algoritmo genético, con un algoritmo de búsqueda local. El resultado es un algoritmo que posee un buen equilibrio entre exploración y explotación.
+
+El procedimiento general del algoritmo queda ilustrado a continuación:
+
+\begin{algorithm}
+    \KwIn{A list $[a_i]$, $i=1, 2, \cdots, n$, that contains the population of individuals}
+    \KwOut{Processed list}
+
+    $P(t) \leftarrow initializePopulation()$
+    $P(t) \leftarrow evaluatePopulation()$
+
+    \While{$\neg stop  condition $}{
+        \If{$certain  iteration$}{
+            $P(t) <- localSearch(P(t-1))$
+        }
+        $t = t + 1$
+
+        $parents \leftarrow selectParents(P(t-1))$
+
+        $offspring \leftarrow recombine(parents)$
+
+        $offspring \leftarrow mutate(offspring)$
+
+        $P(t) \leftarrow replacePopulation(P(t-1), offspring)$
+
+        $P(t) \leftarrow evaluatePopulation()$
+    }
+    \KwRet{$P(t)$}
+\end{algorithm}
+
+Procedemos a la implementación de 3 variantes distintas:
+- Búsqueda local sobre todos los cromosomas
+- Búsqueda local sobre un subconjunto aleatorio de cromosomas
+- Búsqueda local sobre un el subconjunto de los mejores cromosomas
+
+** Implementación
+
+La práctica ha sido implementada en /Python/, usando las siguientes bibliotecas:
+
+- NumPy
+- Pandas
+
+*** Instalación
+
+Para ejecutar el programa es preciso instalar Python, junto con las bibliotecas *Pandas* y *NumPy*.
+
+Se proporciona el archivo shell.nix para facilitar la instalación de las dependencias, con el gestor de paquetes [[https://nixos.org/][Nix]]. Tras instalar la herramienta Nix, únicamente habría que ejecutar el siguiente comando en la raíz del proyecto:
+
+#+begin_src shell
+nix-shell
+#+end_src
+
+** Ejecución
+
+La ejecución del programa se realiza mediante el siguiente comando:
+
+#+begin_src shell
+python src/main.py <dataset> <algoritmo> <parámetros>
+#+end_src
+
+Los parámetros posibles son:
+
+| dataset                              | algoritmo | parámetros                             |
+| Cualquier archivo de la carpeta data | genetic   | uniform/position generation/stationary |
+|                                      | memetic   | all/random/best                        |
+
+También se proporciona un script que ejecuta 1 iteración de cada algoritmo, sobre cada uno de los /datasets/, y guarda los resultados en una hoja de cálculo. Se puede ejecutar mediante el siguiente comando:
+
+#+begin_src shell
+python src/execution.py
+#+end_src
+
+*Nota*: se precisa instalar la biblioteca [[https://xlsxwriter.readthedocs.io/][XlsxWriter]] para la exportación de los resultados a un archivo Excel.
+
+* Análisis de los resultados
+
+Desafortunadamente, debido a un tiempo de ejecución excesivamente alto (incluso tras ajustar los metaparámetros) no podemos proporcionar resultados de la ejecución de los algoritmos.

src/execution.py

@@ -14,70 +14,67 @@ def file_list(path):
 
 
 def create_dataframes():
-    greedy = DataFrame()
+    return [DataFrame() for _ in range(7)]
-    local = DataFrame()
-    return greedy, local
 
 
 def process_output(results):
     distances = []
     time = []
-    for element in results:
+    for line in results:
-        for line in element:
+        if line.startswith(bytes("Total distance:", encoding="utf-8")):
-            if line.startswith(bytes("Total distance:", encoding="utf-8")):
+            line_elements = line.split(sep=bytes(":", encoding="utf-8"))
-                line_elements = line.split(sep=bytes(":", encoding="utf-8"))
+            distances.append(float(line_elements[1]))
-                distances.append(float(line_elements[1]))
+        if line.startswith(bytes("Execution time:", encoding="utf-8")):
-            if line.startswith(bytes("Execution time:", encoding="utf-8")):
+            line_elements = line.split(sep=bytes(":", encoding="utf-8"))
-                line_elements = line.split(sep=bytes(":", encoding="utf-8"))
+            time.append(float(line_elements[1]))
-                time.append(float(line_elements[1]))
     return distances, time
 
 
-def populate_dataframes(greedy, local, greedy_list, local_list, dataset):
+def populate_dataframe(df, output_cmd, dataset):
-    greedy_distances, greedy_time = process_output(greedy_list)
+    distances, time = process_output(output_cmd)
-    local_distances, local_time = process_output(local_list)
+    data_dict = {
-    greedy_dict = {
         "dataset": dataset.removeprefix("data/"),
-        "media distancia": mean(greedy_distances),
+        "media distancia": mean(distances),
-        "desviacion distancia": std(greedy_distances),
+        "desviacion distancia": std(distances),
-        "media tiempo": mean(greedy_time),
+        "media tiempo": mean(time),
-        "desviacion tiempo": std(greedy_time),
+        "desviacion tiempo": std(time),
     }
-    local_dict = {
+    df = df.append(data_dict, ignore_index=True)
-        "dataset": dataset.removeprefix("data/"),
+    return df
-        "media distancia": mean(local_distances),
-        "desviacion distancia": std(local_distances),
-        "media tiempo": mean(local_time),
-        "desviacion tiempo": std(local_time),
-    }
-    greedy = greedy.append(greedy_dict, ignore_index=True)
-    local = local.append(local_dict, ignore_index=True)
-    return greedy, local
 
 
-def script_execution(filenames, greedy, local, iterations=3):
+def script_execution(filenames, df_list):
     script = "src/main.py"
+    parameters = [
+        ["genetic", "uniform", "generational"],
+        ["genetic", "position", "generational"],
+        ["genetic", "uniform", "stationary"],
+        ["genetic", "position", "stationary"],
+        ["memetic", "all"],
+        ["memetic", "random"],
+        ["memetic", "best"],
+    ]
     for dataset in filenames:
         print(f"Running on dataset {dataset}")
-        greedy_list = []
+        for index, params in zip(range(4), parameters):
-        local_list = []
+            print(f"Running {params} algorithm")
-        for _ in range(iterations):
+            output_cmd = run(
-            greedy_cmd = run(
+                [executable, script, dataset, *params], capture_output=True
-                [executable, script, dataset, "greedy"], capture_output=True
             ).stdout.splitlines()
-            local_cmd = run(
+            df_list[index] = populate_dataframe(df_list[index], output_cmd, dataset)
-                [executable, script, dataset, "local"], capture_output=True
+    return df_list
-            ).stdout.splitlines()
-            greedy_list.append(greedy_cmd)
-            local_list.append(local_cmd)
-        greedy, local = populate_dataframes(
-            greedy, local, greedy_list, local_list, dataset
-        )
-    return greedy, local
 
 
-def export_results(greedy, local):
+def export_results(df_list):
-    dataframes = {"Greedy": greedy, "Local search": local}
+    dataframes = {
+        "Generational uniform genetic": df_list[0],
+        "Generational position genetic": df_list[1],
+        "Stationary uniform genetic": df_list[2],
+        "Stationary position genetic": df_list[3],
+        "All genes memetic": df_list[4],
+        "Random genes memetic": df_list[5],
+        "Best genes memetic": df_list[6],
+    }
     writer = ExcelWriter(path="docs/algorithm-results.xlsx", engine="xlsxwriter")
     for name, df in dataframes.items():
         df.to_excel(writer, sheet_name=name, index=False)
@@ -91,9 +88,9 @@ def export_results(greedy, local):
 
 def main():
     datasets = file_list(path="data/*.txt")
-    greedy, local = create_dataframes()
+    df_list = create_dataframes()
-    populated_greedy, populated_local = script_execution(datasets, greedy, local)
+    populated_df_list = script_execution(datasets, df_list)
-    export_results(populated_greedy, populated_local)
+    export_results(populated_df_list)
 
 
 if __name__ == "__main__":

src/local_search.py

@@ -0,0 +1,64 @@
+from numpy.random import choice, seed, randint
+from pandas import DataFrame
+
+
+def get_row_distance(source, destination, data):
+    row = data.query(
+        """(source == @source and destination == @destination) or \
+        (source == @destination and destination == @source)"""
+    )
+    return row["distance"].values[0]
+
+
+def compute_distance(element, solution, data):
+    accumulator = 0
+    distinct_elements = solution.query(f"point != {element}")
+    for _, item in distinct_elements.iterrows():
+        accumulator += get_row_distance(
+            source=element,
+            destination=item.point,
+            data=data,
+        )
+    return accumulator
+
+
+def element_in_dataframe(solution, element):
+    duplicates = solution.query(f"point == {element}")
+    return not duplicates.empty
+
+
+def replace_worst_element(previous, n, data):
+    solution = previous.copy()
+    worst_index = solution["distance"].astype(float).idxmin()
+    random_element = randint(n)
+    while element_in_dataframe(solution=solution, element=random_element):
+        random_element = randint(n)
+    solution["point"].loc[worst_index] = random_element
+    solution["distance"].loc[worst_index] = compute_distance(
+        element=solution["point"].loc[worst_index], solution=solution, data=data
+    )
+    return solution
+
+
+def get_random_solution(previous, n, data):
+    solution = replace_worst_element(previous, n, data)
+    while solution["distance"].sum() <= previous["distance"].sum():
+        solution = replace_worst_element(previous=solution, n=n, data=data)
+    return solution
+
+
+def explore_neighbourhood(element, n, data, max_iterations=100000):
+    neighbourhood = []
+    neighbourhood.append(element)
+    for _ in range(max_iterations):
+        previous_solution = neighbourhood[-1]
+        neighbour = get_random_solution(previous=previous_solution, n=n, data=data)
+        neighbourhood.append(neighbour)
+    return neighbour
+
+
+def local_search(first_solution, n, data):
+    best_solution = explore_neighbourhood(
+        element=first_solution, n=n, data=data, max_iterations=5
+    )
+    return best_solution

src/memetic_algorithm.py

@@ -1,9 +1,46 @@
 from genetic_algorithm import *
 from local_search import local_search
+from copy import deepcopy
 
 
-def run_local_search(n, m, data, individual):
+def get_best_indices(n, population):
-    pass
+    select_population = deepcopy(population)
+    best_elements = []
+    for _ in range(n):
+        best_index, _ = get_best_elements(select_population)
+        best_elements.append(best_index)
+        select_population.pop(best_index)
+    return best_elements
+
+
+def replace_elements(current_population, new_population, indices):
+    for item in indices:
+        current_population[item] = new_population[item]
+    return current_population
+
+
+def run_local_search(n, data, population, mode, probability=0.1):
+    neighbourhood = []
+    if mode == "all":
+        for individual in population:
+            neighbourhood.append(local_search(individual, n, data))
+        new_population = neighbourhood
+    elif mode == "random":
+        expected_individuals = len(population) * probability
+        indices = []
+        for _ in range(expected_individuals):
+            random_index = randint(len(population))
+            random_individual = population[random_index]
+            neighbourhood.append(local_search(random_individual, n, data))
+            indices.append(random_index)
+        new_population = replace_elements(population, neighbourhood, indices)
+    else:
+        expected_individuals = len(population) * probability
+        best_indices = get_best_indices(n=expected_individuals, population=population)
+        for element in best_indices:
+            neighbourhood.append(local_search(population[element], n, data))
+        new_population = replace_elements(population, neighbourhood, best_indices)
+    return new_population
 
 
 def memetic_algorithm(n, m, data, hybridation, max_iterations=100000):
@@ -11,10 +48,10 @@ def memetic_algorithm(n, m, data, hybridation, max_iterations=100000):
     population = evaluate_population(population, data)
     for i in range(max_iterations):
         if i % 10 == 0:
-            best_index, _ = get_best_elements(population)
+            population = run_local_search(n, data, population, mode=hybridation)
-            run_local_search(n, m, data, individual=population[best_index])
+            i += 5
         parents = select_parents(population, n, mode="stationary")
-        offspring = crossover(mode="uniform", parents=parents, m=m)
+        offspring = crossover(mode="position", parents=parents, m=m)
         offspring = mutate(offspring, n, data)
         population = replace_population(population, offspring, mode="stationary")
         population = evaluate_population(population, data)