gurobi / modeling-examples Goto Github PK

Hello,
This is a really great and detailed example of how to use gurobi to solve fantasy sports problems. I actually play draftkings , and have been trying to teach myself how to optimize using gurobi opti... Anyway, I have been attempting to use PoolSolutions to return the other solutions... example:

m.setObjective(obj, sense= GRB.MAXIMIZE)
m.setParam(GRB.Param.PoolSolutions, 500) ## finds 500 best lineups
m.setParam(GRB.Param.PoolSearchMode, 2) ## makes sure these are the best solutions (2)
m.update()

m.optimize()
nSolutions = m.SolCount
print('Number of solutions found: ' + str(nSolutions))
for e in range(nSolutions):
    m.setParam(GRB.Param.SolutionNumber, e)

    for v in m.getVars():
        if v.xN ==1:   ### Help I'm stuck

..... How would I modify your colab notebook to print and save all 500 lineups. I have tried for hours, but just don't have a deep enough understanding of the syntax with gurobi and also creating the correct df calls for this. This isn't an issue, as much as a question . Thanks so much in advance, as I know there's no benefit to you helping me. Again, great work and thank you for sharing on github.

In tsp.ipynb, when excuting the "Model Code" part, you may get a RuntimeError: dictionary changed size during iteration error on

for i, j in vars.keys():
    vars[j, i] = vars[i, j]  # edge in opposite direction

This is because in the "Data computation" part, when creating variable dist, the following code

dist = {(c1, c2): distance(c1, c2) for c1, c2 in combinations(capitals, 2)}

creates a n(n-1)/2 length dictionary, which means variable dist only contains distance[i,j] and does not contain distance[j,i],
for example ,it only contains dist[('Montgomery', 'Phoenix')] but does not contain dist[('Phoenix',Montgomery')]
This leads to further problem when creating and initializing decision variables for gurobi model:

m = gp.Model()

# Variables: is city 'i' adjacent to city 'j' on the tour?
vars = m.addVars(dist.keys(), obj=dist, vtype=GRB.BINARY, name='x')

# Symmetric direction: Copy the object
for i, j in vars.keys():
    vars[j, i] = vars[i, j]  # edge in opposite direction

As stated above, distance[j,i] does not exist, so vars[j, i] does not exist neither. For example, decision variable vars[('Montgomery', 'Phoenix')] exist but vars[('Phoenix',Montgomery')] does not exist, which makes vars[j, i] = vars[i, j] a RuntimeError.

So to fix this , the code
dist = {(c1, c2): distance(c1, c2) for c1, c2 in combinations(capitals, 2)}
should be changed to
dist = {(c1, c2): distance(c1, c2) for c1, c2 in permutations(capitals, 2)}
And of course you need to import itertools.combinations before using it.

@yurchisin

Hi everyone!
I'm modeling with Gurobi, and I'm trying to model an efficiency analysis using different sets of data.
I can't solve the model because it says: "ValueError: not enough values to unpack (expected 3, got 2)".
I attach my model, can anyone understand why it doesn't work?
Thank you in advance!
Matteo

` import pandas as pd
from itertools import product

import gurobipy as gp
from gurobipy import GRB

def solve_DEA(target, verbose=True):
# input-output values for the garages
inattr = ['Cash', 'LEV']
outattr = ['EPS', 'ROA']
dmus,inputs, outputs = gp.multidict({
'DMU1':[{'Cash':0.2485, 'LEV':0.4688 , 'EPS':4, 'ROA':0.2002 }], 'DMU2':[{'Cash':0.1284, 'LEV':0.6539 , 'EPS':3.70, 'ROA':0.0902 }], 'DMU3':[{'Cash':0.293, 'LEV':0.5644 , 'EPS':4.20, 'ROA':0.1627 }]

})

### Create LP model
model = gp.Model('DEA')

# Decision variables
wout = model.addVars(outattr, name="outputWeight")
win = model.addVars(inattr, name="inputWeight")


# Constraints
ratios = model.addConstrs( ( gp.quicksum(outputs[h][r]*wout[r] for r in outattr ) 
                            - gp.quicksum(inputs[h][i]*win[i] for i in inattr ) 
                            <= 0 for h in dmus ), name='ratios' )

normalization = model.addConstr((gp.quicksum(inputs[target][i]*win[i] for i in inattr ) == 1 ),
                                name='normalization')

# Objective function

model.setObjective( gp.quicksum(outputs[target][r]*wout[r] for r in outattr ), GRB.MAXIMIZE)

# Run optimization engine
if not verbose:
    model.params.OutputFlag = 0
model.optimize()

# Print results
print(f"\nThe efficiency of target DMU {target} is {round(model.objVal,3)}") 

print("__________________________________________________________________")
print(f"The weights for the inputs are:")
for i in inattr:
    print(f"For {i}: {round(win[i].x,3)} ") 
    
print("__________________________________________________________________")
print(f"The weights for the outputs are")
for r in outattr:
    print(f"For {r} is: {round(wout[r].x,3)} ") 
print("__________________________________________________________________\n\n")  

return model.objVal

dmus = ['DMU1', 'DMU2', 'DMU3']

performance = {}
for h in dmus:
performance[h] = solve_DEA(h, verbose=False)_
`

In this example, we state that the full dataset can simply be read by replacing small_plant_data by large_plant_data. However, in that folder plant_capacities.csv is missing and the file plant_info_update.csv does not have the same structure so can't serve as an immediate replacement either. What is needed to import the large dataset?

Referring to this notebook: https://gurobi.github.io/modeling-examples/curve_fitting/curve_fitting.html

x should be the independent variable and y should be the dependent variable based on the equation y = ax + b.

RSS vs MSE

This paragraph mentioned that it is not advisable to use RSS as the performance metric, but MSE via cross-validation.

I think the highlight on MSE over RSS is misleading. Note that, given estimate $\hat\beta$,

$$ \mathrm{RSS} = (y-X\hat\beta)^T(y-X\hat\beta) = \sum (y_i - \hat{y}_i)^2 = n \cdot \mathrm{MSE} $$

So, both the training MSE and RSS decrease monotonically as more features are considered, not only RSS.

Cross-validation

The cross-validation part should be the correct. That is, we use grid search to find best $s$.

Dear Sir/Madam,

I am trying to write the constraint related to multiple variables. However, it shows error "Indicator constraints can only be triggered by a single binary variable at a given value". Can you help me fix this error? Thank you very much!
The code is following:
mdl.addConstrs((x[i,j,k] - t[i,j,k] == 1) >> (d2[j,k] == d2[i,k] - d[i,j]) for i, j, k in arcos2 if i != 0 and j != 0)

Where:

• x[i,j,k], t[i,j,k] are binary variables
• d2[j,k], d2[i,k] are continuous variables
• d[i,j] is a parameter

Example https://github.com/Gurobi/modeling-examples/tree/master/technician_routing_scheduling uses the xlrd package. Unfortunately this package does not support reading XLSX files anymore. This may apply to other examples as well.

Hi there, thanks for this great example which I am following - it is really good!

In the model formulation, we have:

I see that we have:
d∈D={n+1,n+2,...,n+m}: Index and set of depots (service centers), where m is the number of depots.
I am confused as to how there can be n+m depots.
I am also confused as to why the set starts from n+1 instead of 1. In the example there are two depots, not 9 i.e 2+7.

I have run into something similar when studying the literature (Ramkumar et al 2012, their formulation also has confusing indices for the number of depots:

)
This makes me think I am missing some key concept in the formulation (as opposed to a typo).
Any enlightenment would be greatly appreciated.

Edit: Indeed, the paper referenced in the notebook also notes "(1,..,n+m) where the m depots are represented by n+1,...,n+m" but does not state their reasoning for doing so. (S. Salhi, A. Imran, N. A. Wassan. The multi-depot vehicle routing problem with heterogeneous vehicle fleet: Formulation and a variable neighborhood search implementation. Computers & Operations Research 52 (2014) 315-325.)

Edit 2: However, in the Supply Network I example, we have something that makes more sense (to me).:

Here there is no (n+1...n+m) indexing.

My objective function is this

\mbox{Average Profit under }(s_1,s_2,...,s_5)=\max \quad & \frac{1}{5}\sum_{k=1}^5 \left(\sum_{i=1}^m\sum_{j=1}^n (12.25-c_{ij})x_{ij}^k\right)\

Here's what I wrote
model.setObjective(quicksum((r-c[i,j]) * math.pow(x[i,j,k] ,k)for i in range(n) for j in range(m) for k in range(5)), GRB.MAXIMIZE)

Errors

-TypeError: must be real number, not Var

Some notebooks have been broken recently by changes in packages, eg

gurobipy 10 -> gurobipy 11
pandas 1 -> pandas 2
gensim not compatible with SciPy 1.13.0

We need to have tight control over the python environment that these notebooks are run under, especially on Google Colab. Unfortunately Colab does not provide a nice way of facilitating this.

I'm proposing we create a public repo modeling-examples-requirements in which we setup a codeless python package with pinned dependencies, essentially creating a lock file in package form. The sole purpose of this package is to tightly control the package versions in an any environment it is installed into. Then in notebooks we only have to have %pip install /path/to/modeling-examples-requirements.git (or something like that).

We could even be fancy and define a separate set of requirements on some different branches to cater for certain notebooks, and avoid having one huge environment. There'd probably be less than 10 different such branches needed.

Hello there!
Using this as base example

I am probably doing something wrong here. I have a distance matrix with costs for traveling from point to point. Not using facility fixed cost, so I want to limit the number of facilities just to fulfill the total demand, such as:

open = m.addVars(list(dict_capacity.keys()), vtype=GRB.BINARY, name="open") #dict capacity has the capacities as values

open_max = m.addVar(vtype=GRB.INTEGER, name="max_open_opportunities", lb=1, ub=32)
m.addConstr(open_max == open.sum())

Assuming that 32 is the max number of facilities I want open, the model takes too long. The way I understood, open.sum() is supposed to give me the sum of the Binary variables on that solution, ie. the number of open facilities. Is it not supposed to have this quicksum quite fast?

m.addConstrs(
            (
                open.sum() * dict_capacity[capacity] <= total_demand + 100
                for capacity in list(dict_capacity.keys())
            )
        )

This is a less hardcoded version I made to test. It works (opens 31 facilities), only because all facilities have 80 units to offer. But it takes so long that I need to Control+C to get objetive value and opened facilities.

open has 628 items

Thanks for your time!

In modeling-examples/optimization101/bike_share/predict_bike_flow.ipynb we try to download a dataset at:

https://s3.amazonaws.com/tripdata/202207-citbike-tripdata.csv.zip

It no longer exists. I'll see if I can find it somewhere else.

In modeling-examples/food_program/food_supply.ipynb we draw a network with nx.nx_pydot.graphviz_layout which requires an external program called Graphviz to be installed.

We can get essentially the same plot with nx.kamada_kawai_layout which is implemented entirely within the networkx package.

Moving to nx.kamada_kawai_layout will make it easier for automated testing.

modeling-examples/music_recommendation/music_recommendation.ipynb relies on lightfm.

The following works in an empty py311 environment:

python -m pip install --upgrade pip setuptools wheel
python -m pip install lightfm

but fails to build in an empty py312 environment. Consequently the notebook can't be used with py312.