JOC / case study code /experiment4.py
BTBUzhy's picture
Upload 4 files
266eb7b verified
Raw History Blame Contribute Delete
20 kB
import numpy as np
import pandas as pd
import math
import random
import warnings
import os
from langchain_openai import ChatOpenAI
import gurobipy as gp
from gurobipy import GRB
import re
import scipy as sp
from langchain_community.utilities import SQLDatabase
from langchain_community.agent_toolkits import create_sql_agent
from langchain_core.runnables import Runnable, RunnableLambda, RunnablePassthrough
from operator import itemgetter
import traceback
from langchain_core.tools import tool
from langchain import PromptTemplate
import time
from langchain_core.messages import AIMessage, HumanMessage
import sys
def add_new_obj(new_model_path, text, w1, w2,unselected,target_variable_values,parameterize,indicator):
import gurobipy as gp
model = gp.Model()
I = list(range(55))
K = [0, 1, 2]
soc = [0, 1, 2]
N = 60
z = [0, 0, 50]
k_0 = 0
k_1 = 1
k_2 = 2
C = [43,36,37,39,44,39,39,39,37,39,36,36,32,42,40,35,43,37,43,37,41,40,37,32,37,31,43,32,44,42,42,31,38,40,41,39,41,40,31,39,40,35,38,39,46,37,41,38,36,41,38,31,43,38,40]
v = np.zeros((55, 3))
random.seed(0)
for i in range(len(I)):
for j in range(len(K)):
v[i, j] = random.randint(j * 5, j * 5 + 10)
for i in range(len(I)):
for k in range(len(K)):
model.addVar(name=f"u_{i}_{k}")
model.addVar(name=f"x_{i}_dispatch_{k}")
for i in range(len(I)):
for k in range(len(K)-1):
model.addVar(name=f"x_{i}_exchange_{k}")
model.update()
exec(text, {'model': model, 'gp': gp, "GRB": gp.GRB, "k_0":k_0, "k_1":k_1, "k_2":k_2, 'I': I, 'K': K, 'N':N, 'z':z, 'v':v, 'C':C})
model.update()
new_obj = model.getObjective()
#print('extra_obj:',new_obj)
new_model = gp.read(new_model_path)
new_model.update()
origin_obj = new_model.getObjective()
new_obj_expr = gp.LinExpr()
if isinstance(new_obj, gp.QuadExpr):
#print('QuadExpr')
w = new_model.addVar(vtype=GRB.CONTINUOUS, name="w")
new_model.update()
new_obj_expr.addTerms(1,w)
new_model.update()
if indicator == 'utilize':
if parameterize is True:
total_expression = gp.QuadExpr()
for i in I:
for k in K:
sum_expression = gp.LinExpr()
var_name = f"u_{i}_{k}"
if i in unselected and var_name in target_variable_values:
#print(var_name,'will be replaced with ',target_variable_values[var_name])
new_model.addConstr(new_model.getVarByName(f'x_{i}_dispatch_{k}') == int(target_variable_values[f'x_{i}_dispatch_{k}']))
new_model.update()
for k in [0,1]:
if i in unselected and var_name in target_variable_values:
new_model.addConstr(new_model.getVarByName(f'x_{i}_exchange_{k}') == int(target_variable_values[f'x_{i}_exchange_{k}']))
new_model.update()
sum_expression = gp.LinExpr()
var_name = f"u_{i}_{k}"
if i in unselected and var_name in target_variable_values:
new_model.update()
sum_expression.addConstant(int(target_variable_values[var_name]))
#total_expression += (target_variable_values[var_name] - C[i])**2
else:
sum_expression.addTerms(1,new_model.getVarByName(f'u_{i}_{k}'))
# print(total_expression)
###
###
###
total_expression += (sum_expression - C[i])**2
###
###
###
else:
###
###
###
total_expression = gp.quicksum((gp.quicksum(new_model.getVarByName(f'u_{i}_{k}') for k in K) - C[i])**2 for i in I)
###
###
###
else:
if parameterize is True:
total_expression = gp.QuadExpr()
for i in I:
for k in [1,2]:
var_name = f"x_{i}_dispatch_{k}"
if i in unselected and var_name in target_variable_values:
#print(var_name,'will be replaced with ',target_variable_values[var_name])
total_expression.addConstant(int(target_variable_values[f"x_{i}_dispatch_{k}"])**2)
else:
total_expression.addTerms(1,new_model.getVarByName(f"x_{i}_dispatch_{k}"),new_model.getVarByName(f"x_{i}_dispatch_{k}"))
for i in I:
for k in K:
if i in unselected and var_name in target_variable_values:
#print(var_name,'will be replaced with ',target_variable_values[var_name])
new_model.addConstr(new_model.getVarByName(f'x_{i}_dispatch_{k}') == int(target_variable_values[f'x_{i}_dispatch_{k}']))
new_model.update()
for k in [0,1]:
if i in unselected and var_name in target_variable_values:
new_model.addConstr(new_model.getVarByName(f'x_{i}_exchange_{k}') == int(target_variable_values[f'x_{i}_exchange_{k}']))
new_model.update()
else:
total_expression = gp.quicksum((new_model.getVarByName(f"x_{i}_dispatch_{k}"))**2 for k in [1,2] for i in I)
new_model.addConstr(w == total_expression,'nonlinear')
new_model.update()
pattern = r'GRB\.(MAXIMIZE|MINIMIZE)'
match = re.search(pattern, str(text))
if match:
new_model.setObjectiveN(w2 * origin_obj, index=0, priority=5, name='obj_tree')
new_model.modelSense = GRB.MINIMIZE
if match.group(1) == 'MAXIMIZE':
new_model.setObjectiveN(-w1 * new_obj_expr,index=1, priority=1, name='obj_indicator')
elif match.group(1) == 'MINIMIZE':
new_model.setObjectiveN(w1 * new_obj_expr,index=1, priority=1, name='obj_indicator')
new_model.update()
#print(new_model.getObjective())
return new_model
def select_random_keys(dictionary, k):
random.seed(int(time.time()))
return random.sample(list(dictionary.keys()), k)
def variables_values(variables, csv_path):
"""
Convert variable names to their generalized form and assign means from a CSV file.
Parameters:
- variables: A list of variable names in string format.
- csv_path: Path to the CSV file.
Returns:
- A dictionary with the original variable names as keys and their calculated means as values.
"""
data = pd.read_csv(csv_path)
result = {}
for var in variables:
# Extract indices from the variable name
indices = re.findall(r'\d+', var)
# Determine the type of variable and build column name for the CSV
if 'exchange' in var:
var_type = 'exchange'
elif 'dispatch' in var:
var_type = 'dispatch'
elif 'u' in var:
var_type = 'I'
if var_type and len(indices) >= 2:
i, k = indices[0], indices[1]
column_name = f"{i}_{var_type}_{k}"
# Calculate the mean value from the CSV data
if column_name in data.columns:
mean_value = int(data[column_name].mean())
result[var] = mean_value
else:
result[var] = None # If the column does not exist in the data
else:
result[var] = None # If not enough indices or unrecognized type
return result
def capture_gurobi_output(log_filename,model):
if os.path.exists(log_filename):
os.remove(log_filename)
model.setParam(GRB.Param.Presolve, 0)
model.setParam(GRB.Param.Cuts, 0)
model.setParam(GRB.Param.TimeLimit, 3600)
#model.setParam(GRB.Param.TimeLimit, 5)
model.setParam(GRB.Param.NonConvex, 2)
model.setParam(GRB.Param.DisplayInterval, 60)
#model.setParam('LogToConsole', 0)
model.setParam('LogFile', log_filename)
start_time = time.time()
model.optimize()
end_time = time.time()
optimize_time = end_time - start_time
var_values = extract_specific_vars_values(model)
with open(log_filename, 'r') as log_file:
log_contents = log_file.read()
best_objective = [float(num) for num in re.findall(r'Best objective ([\d\.e\+\-]+)', log_contents)]
best_bound = [float(num) for num in re.findall(r'best bound ([\d\.e\+\-]+)', log_contents)]
gap = [float(num) for num in re.findall(r'gap ([\d\.e\+\-]+)%', log_contents)]
return optimize_time,best_objective,best_bound,gap,var_values
def evaluate_expression(model, expression,selected):
I = list(range(55))
K = [0, 1, 2]
soc = [0, 1, 2]
N = 60
z = [0, 0, 50]
k_0 = 0
k_1 = 1
k_2 = 2
C = [43,36,37,39,44]
#C = [43,36,37,39,44,39,39,39,37,39,36,36,32,42,40,35,43,37,43,37,41,40,37,32,37,31,43,32,44,42,42,31,38,40,41,39,41,40,31,39,40,35,38,39,46,37,41,38,36,41,38,31,43,38,40]
v = np.zeros((55, 3))
random.seed(0)
for i in range(len(I)):
for j in range(len(K)):
v[i, j] = random.randint(j * 5, j * 5 + 10)
expression = re.sub(r'gp\.quicksum', 'sum', expression)
expression = re.sub(r'model.getVarByName\((.*?)\)', r'model.getVarByName(\1).X', expression)
expression = re.sub(r'I', str(selected), expression, count=1)
expression = re.sub(r'model.setObjective\((.*?), GRB.(MINIMIZE|MAXIMIZE)\)', r'\1', expression)
print(expression)
total_value = eval(expression,{'model': model, 'gp': gp, "GRB": gp.GRB, "k_0":k_0, "k_1":k_1, "k_2":k_2, 'I': I, 'K': K, 'N':N, 'z':z, 'v':v, 'C':C})
print(expression)
return total_value
def extract_specific_vars_values(model):
"""
Extract variable values for variables whose names contain 'x' or 'u' from a solved Gurobi model.
Parameters:
model (gurobipy.Model): The Gurobi model after it has been optimized.
Returns:
dict: A dictionary with variable names as keys and their solution values as values.
"""
var_values = {}
for var in model.getVars():
if 'x' in var.VarName or 'u' in var.VarName:
var_values[var.VarName] = var.X # X attribute gives the solution value of the variable
return var_values
def his_score(csv_path,indicator):
I = list(range(55))
K = [0, 1, 2]
soc = [0, 1, 2]
N = 60
z = [0, 0, 50]
k_0 = 0
k_1 = 1
k_2 = 2
C = [43,36,37,39,44,39,39,39,37,39,36,36,32,42,40,35,43,37,43,37,41,40,37,32,37,31,43,32,44,42,42,31,38,40,41,39,41,40,31,39,40,35,38,39,46,37,41,38,36,41,38,31,43,38,40]
v = np.zeros((55, 3))
random.seed(0)
for i in range(len(I)):
for j in range(len(K)):
v[i, j] = random.randint(j * 5, j * 5 + 10)
data = pd.read_csv(csv_path)
data = data.rename(columns={f"{i}_I_{k}": f"u_{i}_{k}" for i in range(55) for k in range(3)})
data = data.rename(columns={f"{i}_exchange_{k}": f"x_{i}_exchange_{k}" for i in range(55) for k in range(2)})
data = data.rename(columns={f"{i}_dispatch_{k}": f"x_{i}_dispatch_{k}" for i in range(55) for k in range(2)})
mean_data = data.mean(axis=0)[7:].to_dict()
#print(mean_data)
if indicator == 'utilize':
###
###
###
score = sum((sum(mean_data[f'u_{i}_{k}'] for k in K) - C[i])**2 for i in I)
###
###
###
else:
score = sum((mean_data[f'x_{i}_dispatch_{k}'])**2 for k in [1,2] for i in I)
#print(score)
return score
if __name__ == "__main__":
indicators = ['utilize','rate']
###!!!personalized!!!###
csv_path = 'train_data_4.csv'
percentage = 0
times = 1
for t in range(0,times):
indicator = indicators[0]
###!!!personalized!!!###
if t % 3 == 0:
percentage += 0.2
print('\n')
print('\n')
print('第',t,'次')
print('\n')
print('\n')
spot_num = 55*round(percentage,1)
spots = {
0: '3rd Ave & Broad St', 1: '2nd Ave & Blanchard St', 2: '6th Ave & Blanchard St', 3: '2nd Ave & Blanchard St',
4: '7th Ave & Union St', 5: 'Union St & 4th Ave', 6: '1st Ave & Marion St', 7: '2nd Ave & Spring St',
8: 'City Hall / 4th Ave & James St', 9: '2nd Ave & Pine St', 10: '12th Ave & E Yesler Way', 11: 'Summit Ave & E Denny Way',
12: 'E Harrison St & Broadway Ave E', 13: 'Summit Ave E & E Republican St', 14: '15th Ave E & E Thomas St',
15: '12th Ave & E Denny Way', 16: 'E Pine St & 16th Ave', 17: 'Cal Anderson Park / 11th Ave & Pine St',
18: 'Harvard Ave & E Pine St', 19: 'Bellevue Ave & E Pine St', 20: '12th Ave & E Mercer St', 21: '9th Ave N & Mercer St',
22: "Children's Hospital / Sandpoint Way NE & 40th Ave NE",
23: 'Fred Hutchinson Cancer Research Center / Fairview Ave N & Ward St',
24: 'E Blaine St & Fairview Ave E', 25: 'Eastlake Ave E & E Allison St',
26: 'Frye Art Museum / Terry Ave & Columbia St', 27: 'Seattle University / E Columbia St & 12th Ave',
28: '6th Ave S & S King St', 29: 'Occidental Park / Occidental Ave S & S Washington St',
30: 'King Street Station Plaza / 2nd Ave Extension S & S Jackson St', 31: 'REI / Yale Ave N & John St',
32: 'Dexter Ave N & Aloha St', 33: 'Republican St & Westlake Ave N', 34: 'PATH / 9th Ave & Westlake Ave',
35: 'Westlake Ave & 6th Ave', 36: 'Pine St & 9th Ave', 37: 'Lake Union Park / Valley St & Boren Ave N',
38: 'Dexter Ave & Denny Way', 39: 'Key Arena / 1st Ave N & Harrison St', 40: 'Terry Ave & Stewart St',
41: 'Mercer St & 9th Ave N', 42: 'Burke-Gilman Trail / NE Blakeley St & 24th Ave NE',
43: 'NE 42nd St & University Way NE', 44: '12th Ave & NE Campus Pkwy', 45: 'NE 47th St & 12th Ave NE',
46: 'UW McCarty Hall / Whitman Ct', 47: 'Burke Museum / E Stevens Way NE & Memorial Way NE',
48: '15th Ave NE & NE 40th St', 49: 'UW Engineering Library / E Stevens Way NE & Jefferson Rd',
50: 'UW Intramural Activities Building', 51: 'UW Magnuson Health Sciences Center Rotunda / Columbia Rd & San Juan Rd',
52: 'Pier 69 / Alaskan Way & Clay St', 53: 'Seattle Aquarium / Alaskan Way S & Elliott Bay Trail',
54: 'Broadway and E Denny Way'
}
selected = select_random_keys(spots, int(spot_num))
print(selected)
unselected_spots = {key: value for key, value in spots.items() if key not in selected}
unselected = []
for key, value in unselected_spots.items():
unselected.append(key)
print(unselected)
target_variable_values = {}
i_range = list(range(55))
k_range = [0, 1, 2]
vars = []
for i in i_range:
for k in k_range:
var_name = f"u_{i}_{k}"
vars.append(var_name)
var_name = f"x_{i}_dispatch_{k}"
vars.append(var_name)
for k in [0,1]:
var_name = f"x_{i}_exchange_{k}"
vars.append(var_name)
target_variable_values = variables_values(vars,csv_path)
if indicator == 'utilize':
###
###
###
text = "model.setObjective(gp.quicksum((gp.quicksum(model.getVarByName(f'u_{i}_{k}') for k in K) - C[i])**2 for i in I), GRB.MAXIMIZE)"
###
###
###
else:
#min \sum_i \sum_{k\in {k1,k2}} (x_ik)^2
text = "model.setObjective(gp.quicksum(model.getVarByName(f'x_{i}_dispatch_{k}')**2 for k in [1,2] for i in I), GRB.MINIMIZE)"
parameterize_model = add_new_obj('ebike_data/model04.mps', text, 1, 1, unselected, target_variable_values,True,indicator)
# parameterize_model.write('exper4_para.lp')
# parameterize_model.write('exper4_para.mps')
tnow = time.time()
log_path = 'result_extra_exper4/log_files/new_LEO_{}_scale{}_{}.txt'.format(indicator,round(percentage,1), tnow)
LEO_optimize_time,LEO_obj_value,LEO_bound,LEO_gap,LEO_var_values = capture_gurobi_output(log_path,parameterize_model)
LEO_qs = evaluate_expression(parameterize_model, text, selected)
print(LEO_optimize_time,LEO_obj_value,LEO_bound,LEO_gap)
parameterize_model.dispose()
base_model = add_new_obj('ebike_data/model04.mps', text, 1, 1, unselected, target_variable_values,False,indicator)
# base_model.write('exper4_base.lp')
# base_model.write('exper4_base.mps')
log_path = 'result_extra_exper4/log_files/new_base_{}_scale{}_{}.txt'.format(indicator,round(percentage,1), tnow)
base_optimize_time,base_obj_value,base_bound,base_gap,base_var_values = capture_gurobi_output(log_path,base_model)
base_qs = evaluate_expression(base_model, text, selected)
print(base_optimize_time,base_obj_value,base_bound,base_gap)
base_model.dispose()
if indicator == 'utilize':
qs = (LEO_qs-base_qs)/(0.01+base_qs)
else:
qs = -1*(LEO_qs-base_qs)/(0.01+base_qs)
result_gap_1 = (abs(abs(LEO_obj_value[0]) - abs(base_obj_value[0])) / abs(0.00001+LEO_obj_value[0]))*100
result_gap_2 = (abs(abs(LEO_obj_value[1]) - abs(base_obj_value[1])) / abs(0.00001+LEO_obj_value[1]))*100
file_path = 'result_extra_exper4/results/new_{}_scale{}_{}.txt'.format(indicator,round(percentage,1), tnow)
f = open(file_path, 'w')
print('LEO_optimize_time:', LEO_optimize_time)
f.write('LEO_optimize_time:{}'.format(LEO_optimize_time))
f.write('\n')
print('LEO_result:', LEO_obj_value)
f.write('LEO_result:{}'.format(LEO_obj_value))
f.write('\n')
print('LEO_bound', LEO_bound)
f.write('LEO_bound:{}'.format(LEO_bound))
f.write('\n')
print('LEO_gap_1', LEO_gap[0])
print('LEO_gap_2', LEO_gap[1])
f.write('LEO_gap_1:{}'.format(LEO_gap[0]))
f.write('\n')
f.write('LEO_gap_2:{}'.format(LEO_gap[1]))
f.write('\n')
f.write('LEO_qs:{}'.format(LEO_qs))
f.write('\n')
f.write('LEO_var_values:{}'.format(LEO_var_values))
f.write('\n')
print('base_optimize_time:', base_optimize_time)
f.write('base_optimize_time:{}'.format(base_optimize_time))
f.write('\n')
print('base_result:', base_obj_value)
f.write('base_result:{}'.format(base_obj_value))
f.write('\n')
print('base_bound', base_bound)
f.write('base_bound:{}'.format(base_bound))
f.write('\n')
print('base_gap_1', base_gap[0])
print('base_gap_2', base_gap[1])
f.write('base_gap_1:{}'.format(base_gap[0]))
f.write('\n')
f.write('base_gap_2:{}'.format(base_gap[1]))
f.write('\n')
f.write('base_qs:{}'.format(base_qs))
f.write('\n')
f.write('base_var_values:{}'.format(base_var_values))
f.write('\n')
print('selected station IDs:', selected)
f.write('selected station IDs:{}'.format(selected))
f.write('\n')
print('result_gap_1', result_gap_1)
print('result_gap_2', result_gap_2)
f.write('result_gap_1:{}'.format(result_gap_1))
f.write('\n')
f.write('result_gap_2:{}'.format(result_gap_2))
f.write('\n')
print('qs:',qs)
f.write('qs:{}'.format(qs))
f.write('\n')
f.close()