我正在使用python词典来为数学模式定义一些变量和参数。
数据框如下:
Service Bill Weight Zone Resi UPS FedEx USPS DHL
1DEA 1 2 N 33.02 9999 9999 9999
2DAM 2 2 N 33.02 9999 9999 9999
我从中定义了一些输入和变量,如下所示:
cost = {}
for carrier in carriers:
for row in df.to_dict('records'):
key = (row['Service'], row['Bill Weight'],
row['Zone'],row['Resi'], carrier)
cost[key] = row[carrier]
services = df['Service'].unique().tolist()
weights = df['Bill Weight'].unique().tolist()
zones = df['Zone'].unique().tolist()
addresses = df['Resi'].unique().tolist()
我唯一有效的分配和费用组合应为:
['1DEA',1,2,'N','UPS']
['1DEA',1,2,'N','FedEx']
['1DEA',1,2,'N','USPS']
['1DEA',1,2,'N','DHL']
['2DAM',2,2,'N','UPS']
['2DAM',2,2,'N','FedEx']
['2DAM',2,2,'N','USPS']
['2DAM',2,2,'N','DHL']
下面是gurobi python的内容,但我实际上只关心通过python构建循环而不是gurobi语法:
方法A:
assign = {}
for carrier in carriers:
for row in df.to_dict('records'):
key = (row['Service'], row['Bill Weight'],
row['Zone'],row['Resi'], carrier)
cost[key] = row[carrier]
obj = quicksum(cost[key]*assign[key] \
for key in assign)
现在,这种方法可以很好地确保变量和参数仅由字典键生成,而不是由服务,权重,区域和地址的所有可能组合生成。但是当我有如下特定约束时,它将无法工作:
m.addConstrs((assign['1DEA', w, z, r, 'UPS']+assign['1DEA', w, z, r, 'USPS']+assign['1DEA', w, z, r, 'USPS 1C']==1\
for i in clients for s in services for w in weights for z in zones for r in addresses),"C02")
方法2:
assign = m.addVars(services, weights, zones, addresses, carriers, name = "Assign", vtype=GRB.BINARY)
obj = quicksum(cost[s, w, z, r, l]*assign[ s, w, z, r, l] \
for s in services for w in weights for z in zones for r in addresses for l in carriers)
这样,我可以轻松编写所有约束,但是它将创建服务,权重,区域,地址,载体的所有组合,这使我的模型出错了。例如['2DAM',1,2,'N','UPS']不是有效的组合。
是否有一种方法可以将对服务,权重,区域,地址,运营商的这种循环限制为仅在成本字典键中定义的组合?
答案 0 :(得分:1)
由于您的数据已经存在于熊猫数据框中,因此可以使用其功能来创建变量和约束。创建带有决策变量的列,然后使用'groupby'和grb.quicksum定义约束。
首先,更多pythonic列名称
df.columns = ['service', 'bill_weight', 'zone', 'resi', 'UPS', 'FedEx', 'USPS', 'DHL']
然后,将数据框重塑为方便的形式。
df1 = (df.set_index(['service', 'bill_weight', 'zone', 'resi']).
rename_axis('carrier', axis=1).stack().to_frame('cost'))
新数据框每个变量将有一行。
cost
service bill_weight zone resi carrier
1DEA 1 2 N UPS 33.02
FedEx 9999.00
USPS 9999.00
DHL 9999.00
2DAM 2 2 N UPS 33.02
FedEx 9999.00
USPS 9999.00
DHL 9999.00
您可以创建变量(并使用将该变量添加到目标中。
df1['assign'] = [m.addVar(name=".".join(map(str, row.Index),
obj=row.cost, vtype='B')
for row in df1.itertuples()]
m.update()
现在,框架将包含决策变量。
cost assign
service bill_weight zone resi carrier
1DEA 1 2 N UPS 33.02 <gurobi.Var 1DEA.1.2.N.UPS>
FedEx 9999.00 <gurobi.Var 1DEA.1.2.N.FedEx>
USPS 9999.00 <gurobi.Var 1DEA.1.2.N.USPS>
DHL 9999.00 <gurobi.Var 1DEA.1.2.N.DHL>
2DAM 2 2 N UPS 33.02 <gurobi.Var 2DAM.2.2.N.UPS>
FedEx 9999.00 <gurobi.Var 2DAM.2.2.N.FedEx>
USPS 9999.00 <gurobi.Var 2DAM.2.2.N.USPS>
DHL 9999.00 <gurobi.Var 2DAM.2.2.N.DHL>
最后,您可以使用pandas groupby在问题中添加约束,例如
lhs = (df1.groupby(level=['service', 'bill_weight',
'zone', 'resi']).assign.apply(grb.quicksum)
single_carrier_constrs = [m.addConstr(l == 1 for l in lhs]