I would like to define a loss function for the CVXPy optimization that minimizes differences from the reference grouped target:

import cvxpy as cp
import pandas as pd

# toy example for demonstration purpose
target = pd.DataFrame(data={'a': ['X', 'X', 'Y', 'Z', 'Z'], 'b': [1]*5})
w = cp.Variable(target.shape[0])
beta = cp.Variable(target.shape[0])
def loss_func(w, beta):
    x = pd.DataFrame(data={'a': target['a'], 'b': w @ beta}).groupby('a')['b'].sum()
    y = target.groupby('a')['b'].sum()
    return cp.norm2(x - y)**2 # <<<<<<<<<<<<<< ValueError: setting an array element with a sequence.

but this gives me the following error

ValueError: setting an array element with a sequence.

What would be the way to cover this use-case using CVXPy?

I would like to define a loss function for the CVXPy optimization that minimizes differences from the reference grouped target:

import cvxpy as cp
import pandas as pd

# toy example for demonstration purpose
target = pd.DataFrame(data={'a': ['X', 'X', 'Y', 'Z', 'Z'], 'b': [1]*5})
w = cp.Variable(target.shape[0])
beta = cp.Variable(target.shape[0])
def loss_func(w, beta):
    x = pd.DataFrame(data={'a': target['a'], 'b': w @ beta}).groupby('a')['b'].sum()
    y = target.groupby('a')['b'].sum()
    return cp.norm2(x - y)**2 # <<<<<<<<<<<<<< ValueError: setting an array element with a sequence.

but this gives me the following error

ValueError: setting an array element with a sequence.

What would be the way to cover this use-case using CVXPy?

• python
• pandas
• cvxpy

1 Answer 1

Based on my understanding, your code calculates the sum of weighted values (w @ beta) grouped by the unique values in column "a". However, since Pandas cannot handle CVXPY variables, this approach results in errors.

The hstack method, on the other hand, uses native CVXPY functions like cp.sum() and cp.hstack(), making it fully compatible and error-free while giving the same result. Therefore, it’s better to use the hstack approach.‍‍‍

for example:

import cvxpy as cp
import pandas as pd

# toy example for demonstration purpose
target = pd.DataFrame(data={"a": ["X", "X", "Y", "Z", "Z"], "b": [1] * 5})
w = cp.Variable(target.shape[0])
beta = cp.Variable(target.shape[0])

def loss_func2(w, beta):
    x = cp.hstack([
        cp.sum(w[target["a"] == group] * beta[target["a"] == group])
        for group in target["a"].unique()
    ])

    y = target.groupby("a")["b"].sum().values
    return cp.norm2(x - y) ** 2