simulator.py

# -*- coding: utf-8 -*-
"""Overall class for running the simulations.

Contains the Simulator class, which holds all of the matching algorithms and calculates the distortion. Also logs the outcomes of the experiments for analysis and
visualization.

"""

import networkx as nx
import numpy as np
import pandas as pd
import instance_generator
import solver
import argparse


#TODO the experiment code in here has a ton of repetition - think about how this could be better formatted?
class Simulator:
    """Object that solves instances of matching problems given to it and aggregates the results in a pretty manner.

    Attributes:
        instance_generator (InstanceGenerator): InstanceGenerator for creating problem instances for simulation
        history (dict): history of all results, stored in a dictionary for later conversion to pd.DataFrame
    """

    def __init__(self, instance_generate):
        """Initializes a new Simulator.

        Args:
            instance_generate (InstanceGenerator): InstanceGenerator for creating problem instances for simulation
        """
        self.instance_generator = instance_generate
        self.history = {'id': [], 'val_index': [], 'size': [], 'valuation':[], 'algo': [], 'distortion': []}
        self.id = 1


    def serial_dictatorship_experiment(self, val_index, val_type, G, size=None, agent_cap=None):
        """Finds the distortion of running serial dictatorship on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
            
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.serial_dictatorship(G,agent_cap)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('serial_dictatorship')
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M,prio='pareto'))


    def partial_max_matching_experiment(self, val_index, val_type, G, m, size=None, agent_cap=None):
        """Finds the distortion of running PartialMaxMatching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            m (int): The number of buckets, used as input to PartialMaxMatching.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
            
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.partial_max_matching(G,m,agent_cap)
        H = solver.reassign_labels(G, M)
        M_0 = solver.top_trading_cycles(H)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('partial_max_matching'+ '_' + str(m))
        self.history['distortion'].append(solver.calculate_distortion(G,M_0))        


    def modified_max_matching_experiment(self, val_index, val_type, G, prio='pareto', size=None, agent_cap=None):
        """Finds the distortion of running ModifiedMaxMatching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
            
        """
        if size is None:
            size = len(G.nodes)//2

        M_0 = solver.modified_max_matching(G,prio=prio,agent_cap=agent_cap)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('modified_max_matching')
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M_0,prio))


    def top_trading_cycles_experiment(self, val_index, val_type, G, size=None, agent_cap=None):
        """Finds the distortion of running top trading cycles on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.top_trading_cycles(G,agent_cap)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('ttc_matching')
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M,prio='pareto'))

    
    def epsilon_max_matching_experiment(self, val_index, val_type, G, epsilon, size=None, agent_cap=None):
        """Finds the distortion of running epsilon max matching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.epsilon_max_matching(G, epsilon, agent_cap=agent_cap)
        H = solver.reassign_labels(G, M)
        M_0 = solver.top_trading_cycles(H)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('epsilon_max_matching'+str(epsilon))
        self.history['distortion'].append(solver.calculate_distortion(G,M_0))


    def epsilon_max_matching_prio_experiment(self, val_index, val_type, G, epsilon, prio='pareto', size=None, agent_cap=None):
        """Finds the distortion of running epsilon max matching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            prio (String): String in ['rank_maximal', 'max_cardinality_rank_maximal', 'fair'] that represents the priority vector used for this problem. Defaults to 'rank_maximal'.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.epsilon_max_matching(G, epsilon, prio, agent_cap)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('epsilon_max_matching '+prio+str(epsilon))
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M,prio))


    #TODO fix inconsistent casing
    def twothirds_max_matching_experiment(self, val_index, val_type, G, prio, size=None, agent_cap=None):
        """Finds the distortion of twothirds_max_matching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            prio (String): String in ['rank_maximal', 'max_cardinality_rank_maximal', 'fair'] that represents the priority vector used for this problem. Defaults to 'rank_maximal'.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
        """
        M = solver.twothirds_max_matching(G, prio, agent_cap)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('twothirds_max_matching '+prio)
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M,prio))

    def updated_hybrid_max_matching_experiment(self, val_index, val_type, G, size=None, agent_cap=None):
        """Finds the distortion of running updated HybridMaxMatching on the given input.
        
        Args:
            val_index (int): The id of the valuation in self.instance_generator.
            val_type (string): The method by which the valuation was generated.
            G (nx.Graph): The actual matching input. Must be a weighted bipartite graph with numerical node labels.
            size (int): The number of agents in the input. Defaults to half the size of G.nodes if not given.
            agent_cap (int): An integer value i such that for all nodes with label <= i, those nodes are agents. Defaults to len(G.nodes//2) if not given.
            
        """
        if size is None:
            size = len(G.nodes)//2

        M = solver.updated_hybrid_max_matching(G,agent_cap=agent_cap)
        H = solver.reassign_labels(G, M)
        M_0 = solver.top_trading_cycles(H)

        self.history['id'].append(self.id)
        self.id += 1
        self.history['val_index'].append(val_index)
        self.history['size'].append(size)
        self.history['valuation'].append(val_type)
        self.history['algo'].append('updated_hybrid_max_matching')
        self.history['distortion'].append(solver.calculate_modified_distortion(G,M_0,prio='pareto'))


if __name__=='__main__':
    instantiator = instance_generator.InstanceGenerator(True)
    sim = Simulator(instantiator)
    # for n in [5,10,20,50,100]: # adjust the number of intervals here
    #     print('current batch is', n) 

    #     for j in range(20): # adjust number of trials per n here

    #         G = instantiator.generate_unit_range_arrow(n, -1) #adjust the valuation generation method here
    #         val_index = instantiator.index-1
    #         val_type = 'unit_range_arrow_-1'

    #         sim.serial_dictatorship_experiment(val_index,val_type,G)
    #         sim.updated_hybrid_max_matching_experiment(val_index,val_type,G)
    #         sim.top_trading_cycles_experiment(val_index,val_type,G)
    #         sim.epsilon_max_matching_experiment(val_index,val_type,G,1)
    #         sim.epsilon_max_matching_experiment(val_index,val_type,G,0.1)

    
    # df = pd.DataFrame(sim.history)
    # df.to_csv("C:/Users/sqshy/Desktop/University/Fifth Year/research/DistortionSim/updateddata/unit_range_arrow_-1_pareto.csv") #adjust path name here

    # df = pd.Series(sim.instance_generator.history)
    # df.to_csv("C:/Users/sqshy/Desktop/University/Fifth Year/research/DistortionSim/updateddata/unit_range_arrow_-1_pareto.csv") #adjust instance data path name here

    #still need to modify which experiments to run from below
    #parameters for directly modifying arguments from file
    theta=0.2
    norm = "range"
    crit = "" # empty string and pareto are the same

    #argparse parameters
    parser = argparse.ArgumentParser("DistortionSim")
    parser.add_argument("--theta", type=str, default="1", choices=["0.2", "1", "5"])
    parser.add_argument("--norm", type=str, default="range", choices=["range", "sum"])
    parser.add_argument("--crit", type=str, default="", choices=["", "pareto", "rank_maximal", "max_cardinality_rank_maximal", "fair"])
    parser.add_argument("--save_dir", type=str, default="./")
    parser.add_argument("--ckpt_path", type=str, default="./")

    args = parser.parse_args()

    theta = args.theta
    norm = args.norm
    crit = args.crit

    #end parameter differences

    val_type = f"theta{theta}unit{norm}"
  

    sizes = [5,10,20,50,100]

    for size in [5,10,20,50,100]:
        filename = f"rdata/ord_n{size}_theta{theta}.txt"
        print('current n value is', size)

        G_list = instantiator.generate_list_from_ordinal_preferences(filename, size, 100, "unit_"+norm) #adjust unit-range vs unit-sum here
        val_index = instantiator.index - 100

        for i in range(len(G_list)):
            G = G_list[i]

            #adjust experiments here
            sim.serial_dictatorship_experiment(val_index,val_type,G)
            sim.top_trading_cycles_experiment(val_index,val_type,G)
            sim.epsilon_max_matching_prio_experiment(val_index,val_type,G,1,prio=crit)
            sim.epsilon_max_matching_prio_experiment(val_index,val_type,G,0.1,prio=crit)
            sim.modified_max_matching_experiment(val_index,val_type,G,prio=crit)
            sim.updated_hybrid_max_matching_experiment(val_index,val_type,G)

            val_index += 1

    # adjust naming conventions here
    s = "ijcaidata/"+val_type+crit+".csv"
    s_instances = "ijcaidata/"+val_type+crit+"instances.csv"

    df = pd.DataFrame(sim.history)
    df.to_csv(s)

    df = pd.Series(sim.instance_generator.history)
    df.to_csv(s_instances)