fuzzy.py

import pandas as pd
import numpy as np
import os
import json
import skfuzzy as fuzz
import skfuzzy.membership as mf
import matplotlib.pyplot as plt

def get_path(starts_with,directory=''):
  directory=os.path.join(os.getcwd(),directory)
  file=[filename for filename in os.listdir(directory) if filename.startswith(starts_with)]
  return os.path.join(directory,file[0])


def get_data():
  '''Return the data needed to calculate payoffs'''
  #get the data
  df_abort=pd.read_csv(get_path('abortion','processed_data'))
  df_ed=pd.read_csv(get_path('educa','processed_data'))
  df_gdp=pd.read_csv(get_path('gdp','processed_data'))
  df_health=pd.read_csv(get_path('health','processed_data'))

  #put it in a better data format
  s_i=np.array(range(len(df_abort)))
  s_code=df_abort['state'].values
  s_name=df_ed['State'].values
  s_abort=df_abort['Average Abortion Rate'].values
  s_ed=df_ed['DegreePerCapita'].values
  s_gdp=df_gdp['Average GDP'].values
  s_health=[]
  for state in s_name:
    spending=df_health[df_health['State']==state]['Average Spending'].values[0]
    s_health.append(spending)
  s_health=np.array(s_health)

  s_code=np.char.lower(s_code.astype(str))
  s_name=np.char.lower(s_name.astype(str))

  #get the swing state data
  with open(get_path('swing','processed_data')) as f:
    swing_dict=json.load(f)
  s_last_party_l=[]
  s_margin_l=[]
  s_left_right_l=[]
  indexes_l=[]
  for state in swing_dict:
    indexes_l.append(np.where(s_name==state[0].lower())[0][0])
    att=state[1]
    s_last_party_l.append(att['last_party'])
    s_margin_l.append(att['average_margin'])
    if att['last_party']=='REPUBLICAN':
      s_left_right_l.append(((1-att['average_margin'])/2,(1+att['average_margin'])/2))
    else:
      s_left_right_l.append(((1+att['average_margin'])/2,(1-att['average_margin'])/2))

  s_last_party=np.array(s_last_party_l)
  s_margin=np.array(s_margin_l)
  s_left_right=np.array(s_left_right_l)

  s_last_party=s_last_party[np.argsort(np.array(indexes_l))]
  s_margin=s_margin[np.argsort(np.array(indexes_l))]
  s_left_right=s_left_right[np.argsort(np.array(indexes_l))]

  #package data for return statement
  strategies=[s_abort,s_ed,s_gdp,s_health]
  small_is_good_for_rep=[True,True,False,True]
  data=dict()
  data['strategies']=strategies
  data['sigfr']=small_is_good_for_rep
  data['sigfd']=[not v for v in small_is_good_for_rep]
  data['s_i']=s_i
  data['s_code']=s_code
  data['s_name']=s_name
  data['s_last_party']=s_last_party
  data['s_margin']=s_margin
  data['s_left_right']=s_left_right

  return data

def get_strategy_quality(state,metric,small_is_good=True):
  value=metric[state]
  if small_is_good:
    sorted=np.sort(metric)[::-1]
  else:
    sorted=np.sort(metric)
  rank=np.where(sorted==value)[0]
  return rank/(len(metric)-1)

def plot(dist,title):
  plt.figure(figsize=(2,2))
  plt.title(title)
  plt.ylim(0,1)
  plt.plot(dist)


def calculate_payoff(dem_s,rep_s,state,data,election_outcome=True,log=False,charts=False):
  '''
  Calculate the payoff for each candidate\n
  If election_outcome=True, returns who wins the state (dem,rep)\n
  If election_outcome=False, returns the voting results (dem,rep)\n
  If log = True prints out logs
  If charts = True prints out graphs
  dem_s,rep_s can take value {0,1,2,3} <-> promote:{abortion,education,gdp,heath}\n
  state can take value {0,1,...,9} <->{0:'colorado', 1:'florida', 2:'georgia',
  3:'iowa', 4:'louisiana', 5:'michigan', 6:'new mexico', 7:'ohio',
  8:'pennsylvania', 9:'wisconsin'}
  '''
  #Fuzzy set of how convincing a strategy is
  x=np.arange(0,11,1)
  convinced=mf.sigmf(x,5,0.3)
  if charts:plot(convinced,'convinced set')
  #how convincing the dem strategy is
  strat_qual=get_strategy_quality(state,data['strategies'][dem_s],data['sigfd'][dem_s])*10
  if log:print('dem_s qual:',strat_qual)
  dem_s_effect=fuzz.interp_membership(x,convinced,strat_qual)
  #how convincing the rep strategy is
  strat_qual=get_strategy_quality(state,data['strategies'][rep_s],data['sigfr'][rep_s])*10
  if log:print('rep_s qual:',strat_qual)
  rep_s_effect=fuzz.interp_membership(x,convinced,strat_qual)
  #People's a priori desire to vote left or right
  vote_left_right=np.arange(0,1001,1)
  vote_left=mf.trapmf(vote_left_right,[0,0,350,800])
  if charts:plot(vote_left,'dem a priori likeability')
  vote_right=mf.trapmf(vote_left_right,[200,650,1000,1000])
  if charts:plot(vote_right,'rep a priori likeability')
  if log:print('dem_s effect:',dem_s_effect)
  if log:print('rep_s effect:',rep_s_effect)
  #People's updated desire to vote left or right after the campaign
  vote_left=np.fmin(dem_s_effect,vote_left)
  if charts:plot(vote_left,'dem posteriori likeability')
  vote_right=np.fmin(rep_s_effect,vote_right)
  if charts:plot(vote_right,'rep posteriori likeability')
  #The combined outcome of the campaigns
  combined_vote_intention=np.fmax(vote_left,vote_right)
  if charts:plot(combined_vote_intention,'combined effect of the campaigns')
  results=fuzz.defuzz(vote_left_right,combined_vote_intention,'centroid')
  results=1-(results/1000)
  #The result of the compaigned is averaged out with previous election results
  # to reflect people's unwillingness to change
  results_left_right=((data['s_left_right'][state][0]+results)/2,(data['s_left_right'][state][1]+1-results)/2)
  if log:print('election results:',results_left_right)

  if election_outcome:
    if results_left_right[0]>results_left_right[1]:
      return (1,0)
    else:
      return (0,1)
  else:
    return results_left_right