Merge branch 'main' into fix-conflict-init

experiments/compare_cache_in_tabu_search.py

@@ -0,0 +1,185 @@
+"""
+Compare the performance of algorithms for fair course allocation.
+
+Programmer: Erel Segal-Halevi
+Since: 2023-07
+"""
+import os
+######### COMMON VARIABLES AND ROUTINES ##########
+
+import random
+
+from fairpyx import Instance, AgentBundleValueMatrix, divide
+from typing import *
+import numpy as np
+import matplotlib.pyplot as plt
+import pandas as pd
+import ast
+import seaborn as sns
+import fairpyx.algorithms as crs
+
+max_value = 1000
+normalized_sum_of_values = 1000
+TIME_LIMIT = 100
+
+
+def create_initial_budgets(num_of_agents: int, beta: float = 100) -> dict:
+    # Create initial budgets for each agent, uniformly distributed in the range [1, 1 + beta]
+    initial_budgets = np.random.uniform(1, 1 + beta, num_of_agents)
+    return {f's{agent + 1}': initial_budgets[agent] for agent in range(num_of_agents)}
+
+
+def evaluate_algorithm_on_instance(algorithm, instance, **kwargs):
+    beta = kwargs.get("beta", 100)
+    initial_budgets = create_initial_budgets(instance.num_of_agents, beta)
+    allocation = divide(algorithm, instance=instance, initial_budgets=initial_budgets, **kwargs)
+
+    matrix = AgentBundleValueMatrix(instance, allocation)
+    matrix.use_normalized_values()
+    return {
+        "utilitarian_value": matrix.utilitarian_value(),
+        "egalitarian_value": matrix.egalitarian_value(),
+        "max_envy": matrix.max_envy(),
+        "mean_envy": matrix.mean_envy(),
+        "max_deficit": matrix.max_deficit(),
+        "mean_deficit": matrix.mean_deficit(),
+        "num_with_top_1": matrix.count_agents_with_top_rank(1),
+        "num_with_top_2": matrix.count_agents_with_top_rank(2),
+        "num_with_top_3": matrix.count_agents_with_top_rank(3),
+    }
+
+
+######### EXPERIMENT WITH UNIFORMLY-RANDOM DATA ##########
+
+def course_allocation_with_random_instance_uniform(
+        num_of_agents: int, num_of_items: int,
+        value_noise_ratio: float,
+        algorithm: Callable,
+        random_seed: int, **kwargs):
+    agent_capacity_bounds = [6, 6]
+    item_capacity_bounds = [40, 40]
+    np.random.seed(random_seed)
+    instance = Instance.random_uniform(
+        num_of_agents=num_of_agents, num_of_items=num_of_items,
+        normalized_sum_of_values=normalized_sum_of_values,
+        agent_capacity_bounds=agent_capacity_bounds,
+        item_capacity_bounds=item_capacity_bounds,
+        item_base_value_bounds=[1, max_value],
+        item_subjective_ratio_bounds=[1 - value_noise_ratio, 1 + value_noise_ratio]
+    )
+    return evaluate_algorithm_on_instance(algorithm, instance, **kwargs)
+
+
+######### COMPARING USING CACHE - TABU SEARCH ##########
+
+RESULTS_CACHE_TABU_SEARCH = "results/compering_using_cache_tabu_search.csv"
+
+def run_cache_experiment_tabu_search():
+    # Run on uniformly-random data with beta and delta parameters:
+    experiment = experiments_csv.Experiment("results/", "compering_using_cache_tabu_search.csv",
+                                            backup_folder="results/backup/")
+    input_ranges = {
+        "num_of_agents": [10, 20, 30],
+        "num_of_items": [5, 10, 15],
+        "value_noise_ratio": [0, 0.2],
+        # "value_noise_ratio": [0, 0.2, 0.4, 0.8, 1],
+        "beta": [3],
+        "delta": [{0.5}],
+        # "delta": [{0.001}, {0.1}, {0.3}, {0.5}, {0.9}],
+        "use_cache": [False, True],
+        "algorithm": [crs.tabu_search],
+        "random_seed": range(5),
+    }
+    experiment.run_with_time_limit(course_allocation_with_random_instance_uniform, input_ranges, time_limit=TIME_LIMIT)
+
+
+
+def analyze_experiment_results_cache():
+    # Load the results from the CSV file
+    df = pd.read_csv(RESULTS_CACHE_TABU_SEARCH)
+
+    best_row = df.loc[df['runtime'].idxmin()]
+
+    # Extract relevant columns or parameters
+    best_use_cache_value = best_row['use_cache']
+    best_runtime = best_row['runtime']
+
+    print(f"Best use_cache: {best_use_cache_value}")
+    print(f"Corresponding Runtime: {best_runtime} seconds")
+
+    return df
+
+
+##### PLOT #######
+
+def plot_speed_vs_param(df, param, algorithm_name):
+    avg_runtime = df.groupby(param)['runtime'].mean().reset_index()
+
+    plt.figure(figsize=(10, 6))
+    sns.lineplot(data=avg_runtime, x=param, y='runtime', marker='o', err_style=None)
+    plt.title(f'Algorithm Speed vs. {param.capitalize()} for {algorithm_name}')
+    plt.xlabel(param.capitalize())
+    plt.ylabel('Average Runtime (seconds)')
+    plt.grid(True)
+    plt.tight_layout()
+    plt.show()
+
+
+def plot_speed_vs_params(df, param1, param2, algorithm_name):
+    fig = plt.figure(figsize=(10, 6))
+    ax = fig.add_subplot(111, projection='3d')
+
+    param1_values = df[param1]
+    param2_values = df[param2].apply(
+        lambda x: float(list(ast.literal_eval(x))[0]) if isinstance(x, str) else x)  # Convert if necessary
+    runtime_values = df['runtime']
+
+    ax.scatter(param1_values, param2_values, runtime_values, c='b', marker='o')
+    ax.set_title(f'Algorithm Speed vs. {param1.capitalize()} and {param2.capitalize()} for {algorithm_name}')
+    ax.set_xlabel(param1.capitalize())
+    ax.set_ylabel(param2.capitalize())
+    ax.set_zlabel('Runtime (seconds)')
+
+    plt.tight_layout()
+    plt.show()
+
+
+def plot_runtime_vs_cache(df, algorithm_name):
+    plt.figure(figsize=(12, 8))
+
+    # Plotting runtime vs. num_of_agents for each use_cache value
+    sns.lineplot(data=df[df['use_cache'] == True], x='num_of_agents', y='runtime', marker='o',
+                 label='Use Cache = True')
+    sns.lineplot(data=df[df['use_cache'] == False], x='num_of_agents', y='runtime', marker='o',
+                 label='Use Cache = False')
+
+    plt.title(f'Runtime vs. Number of Agents for {algorithm_name}')
+    plt.xlabel('Number of Agents')
+    plt.ylabel('Runtime (seconds)')
+    plt.grid(True)
+
+    # Customizing the legend
+    legend_labels = {
+        'Use Cache = True': 'blue',  # Blue line and markers for Use Cache = True
+        'Use Cache = False': 'orange'  # Orange line and markers for Use Cache = False
+    }
+
+    handles = [plt.Line2D([0, 0], [0, 0], color=color, marker='o', linestyle='') for color in legend_labels.values()]
+    labels = legend_labels.keys()
+    plt.legend(handles, labels, title='Legend')
+
+    plt.tight_layout()
+    plt.show()
+
+
+
+###########################
+
+if __name__ == "__main__":
+    import logging, experiments_csv
+
+    experiments_csv.logger.setLevel(logging.INFO)
+
+    run_cache_experiment_tabu_search()
+    df = analyze_experiment_results_cache()
+    plot_runtime_vs_cache(df, 'Tabu Search')

experiments/compare_course_allocation_algorithms.py

@@ -166,7 +166,7 @@ def run_ariel_experiment():
     import logging, experiments_csv
     experiments_csv.logger.setLevel(logging.INFO)
     run_uniform_experiment()
-    run_szws_experiment()
-    run_ariel_experiment()
-
+    # run_szws_experiment()
+    # run_ariel_experiment()
+    #