Python pulp error when exporting a problem to json file.

[moheikal79]

Hi,
Please for advice, I am facing a problem when exporting a python pulp problem to json file "-75 is not JSON serializable".

variables= pulp.LpVariable.dicts("x", (item for item in df.index), lowBound=0, cat='Integer')

model = pulp.LpProblem("Nova_2_levels_maximize", LpMaximize)

model += pulp.lpSum([variables[item] * df.loc[item, 'Profitability'] for item in df.index])

max_weight = 500000
model += pulp.lpSum([variables[item] * df.loc[item, 'Weight'] for item in df.index]) <= max_weight

max_size = 100000
model += pulp.lpSum([variables[item] * df.loc[item, 'Size'] for item in df.index]) <= max_size

min_sales = 4000
model += pulp.lpSum([variables[item] * df.loc[item, 'Sales price'] for item in df.index]) >= min_sales

min_qty = df.index
for nq in min_qty:
model += variables[nq] >= df.loc[nq, 'Min Qty sales']

max_qty = df.index
for xq in max_qty:
model += variables[xq] <= df.loc[xq, 'Max Qty sales']

model.solve()

model.to_json(r"D:\PL\test.json")

[pchtsp]

That's interesting. It's possible there's an error in export function (or your model has something we did not expect). Do you have a reproducible example we can test to find the source of the problem?

[moheikal79]

Thanks for your reply, I am attaching the excel version of the data use and the code as well.

Regards

Nova LP.xlsx
pulp+json.txt

[pchtsp]

Hello again,

I have checked your code. Apparently, the python json module does not like some of the pandas / numpy data types (floats, int64, etc.).

You have to choices:

option1: type casting, e.g.:

#(...)
for xq in max_qty:
    model += variables[xq] <= int(df.loc[xq, 'Max Qty sales'])
# you need to change it every time you use a pandas data.frame, using `float` or `int` at each time
#(...)

option2: pass a custom encoder to the model.toJson function as explained here:

import numpy as np
#(...)
class NpEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.integer):
            return int(obj)
        elif isinstance(obj, np.floating):
            return float(obj)
        elif isinstance(obj, np.ndarray):
            return obj.tolist()
        else:
            return super(NpEncoder, self).default(obj)

model.to_json(r"D:\PL\test.json", cls=NpEncoder)

Given that many people actually use pandas to formulate problems for pulp, I think I will update the docs with this information.

[moheikal79]

Thanks for reply, I have used option 1 and added int() to all constraints and the step of adding variables to the model, and it works fine. It will be a great step to include examples of using pandas along with pulp as they are used together in production. I am attaching the code after resolving the issue.

pulp+json.txt

Regards

[pchtsp]

In fact, I do not like using pandas with pulp. Nevertheless, I know many people use it since it's easy to pass from tables to model. I think we had envisioned a "best practices" page somewhere in the docs. We should one day put something about that. F.

…

On Tue, Mar 30, 2021, 18:58 moheikal79 ***@***.***> wrote: Thanks for reply, I have used option 1 and added int() to all constraints and the step of adding variables to the model, and it works fine. It will be a great step to include examples of using pandas along with pulp as they are used together in production. I am attaching the code after resolving the issue. pulp+json.txt <https://github.com/coin-or/pulp/files/6230785/pulp%2Bjson.txt> Regards — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#419 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABJUZ46XFB4RCNPEP2J5FQLTGH7MJANCNFSM4Z5AWO4A> .