I am trying to create a modified implementation of fused lasso in cvxpy to solve an optimization problem. The following code gives me an error of:
DCPError: Problem does not follow DCP rules. Specifically:
The objective is not DCP. Its following subexpressions are not:
max(Vstack(var2111[0:190], var2111[190:380], var2111[380:570], var2111[570:760]), 0, False)[1:190] + -max(Vstack(var2111[0:190], var2111[190:380], var2111[380:570], var2111[570:760]), 0, False)[0:189]
def loss_fn(X, Y, beta_A, beta_C, beta_G, beta_T):
return cp.norm2(Y - [email protected]((beta_A, beta_C, beta_G, beta_T)))**2
def regularizer(beta_A, beta_C, beta_G, beta_T):
beta_max = cp.max(cp.vstack((beta_A, beta_C, beta_G, beta_T)), axis = 0)
print (beta_max)
return cp.sum(cp.abs(cp.diff(beta_max)))
def objective_fn(X, Y, beta_A, beta_C, beta_G, beta_T, lambd):
return loss_fn(X, Y, beta_A, beta_C, beta_G, beta_T) + lambd *
regularizer(beta_A, beta_C, beta_G, beta_T)
beta_A = cp.Variable(len(seq))
beta_C = cp.Variable(len(seq))
beta_G = cp.Variable(len(seq))
beta_T = cp.Variable(len(seq))
lambd = cp.Parameter(nonneg=True)
problem = cp.Problem(cp.Minimize(objective_fn(X_diff_xor.values, y_values,
beta_A, beta_C, beta_G, beta_T, lambd)))
#lambd_values = [3, 5, 7, 9, 11, 13] # regularization parameter
lambd_values = [7]
beta_values = []
for v1 in lambd_values:
lambd.value = v1
problem.solve(solver = 'ECOS', verbose = True)
beta_values.append(beta.value)
Any help/insight will be appreciated. Thanks.
question from:
https://stackoverflow.com/questions/65889746/cvxpy-dcperror-problem-does-not-follow-dcp-rules