cvxpy + mosek: The lower bound specified is not a number (nan)

Question

In below code snippet I am trying to do a very simple linear regression using cvxpy and mosek solver

weight = df2['weight'].to_numpy()
A = df2[ regressors ].to_numpy()
x = cp.Variable(len(regressors ));
R = df2['y'].to_numpy()
cost = cp.sum_squares( A @ x - R)
print(A.shape, x.shape, R.shape)          # here it prints (134882, 8) (8,) (134882,)
prob = cp.Problem(cp.Minimize(cost), 
           [x >= 0])

but I keep getting below error.

Error: rescode.err_lower_bound_is_a_nan(1390): The lower bound specified is not a number (nan).

I have checked the manual: https://docs.mosek.com/latest/pythonapi/response-codes.html#mosek.rescode.err_lower_bound_is_a_nan but what exactly am I missing?

Your input data contain NANs is the best bet. Or a bug in cvxpy. Btw why do you minimize sum of squares? Minimizing the norm is better. — ErlingMOSEK, Sep 11 '21 at 05:21
@ErlingMOSEK forive my ignorance, why is it that minimizing the norm is better than minimize sum of squares? — Matt Frank, Sep 11 '21 at 20:05
The norm behaves like a linear function ond is better scaled. In any case cvxpy will convert the problem into s norm problem before sending it to Mosek. Mosek prefers the norm form. — ErlingMOSEK, Sep 12 '21 at 05:02

score 0 · Accepted Answer · answered Sep 10 '21 at 19:25

0

it's hard to answer this question without having a reproducible example with your data.

Your python code seems fine, so I believe the problem is in your data: if I must take a guess, I assume you have NaNs somewhere in your data which causes the failure.

I'm not sure if cvxpy and mosek handle nans at all, so it's a good idea to verify your data integrity before using them.

answered Sep 10 '21 at 19:25

Roim

2,986
2
10
25

hah indeed.... there is a nan. i wish they give a more clear error message. – Matt Frank Sep 10 '21 at 20:10
Cvxpy should check for NANs if a better message should be produced. – ErlingMOSEK Sep 13 '21 at 08:07

cvxpy + mosek: The lower bound specified is not a number (nan)

1 Answers1